Superbus Positioning System · Hiermee wordt duidelijk dat een replicator (genoom) een ratio heeft boven die van het individu (replicant). Je kunt niet anders dan zeggen dat het genoom

Gert-Jan Pauwels

Superbus Positioning SystemA High Accuracy Networked RTK GPS System

Delft, 2010

Master’s Thesis

Superbus Positioning SystemA High Accuracy Networked RTK GPS System

MASTER OF SCIENCE THESIS

For obtaining the degree of Master of Science in AerospaceEngineering at Delft University of Technology

Gert-Jan Pauwels1097407

March 7, 2011

Faculty of Aerospace Engineering · Delft University of Technology

Copyright c© Gert-Jan PauwelsAll rights reserved.

DELFT UNIVERSITY OF TECHNOLOGYCHAIR OF

MATHEMATICAL GEODESY AND POSITIONING (MGP)

The undersigned hereby certify that they have read and recommend to the Facultyof Aerospace Engineering for acceptance a thesis entitled “Superbus Positioning Sys-tem” by Gert-Jan Pauwels in partial fulfillment of the requirements for the degree ofMaster of Science.

Dated: March 7, 2011

Head of chair MGP:prof.dr.ir. R.F. Hanssen

Supervisor:dr.ir. Christian Tiberius

Supervisor:Dr. Antonia Terzi

Acknowledgements

The road I took during the making of this thesis, has had me thinking frequentlyof the tortoise and the hare parable. I have come to the conclusion that a third creatureshould’ve entered the race. I propose a creature not necessarily as slow as a tortoise,however with a horrible sense of direction. Eventually stumbling across the finishline, having completely forgotten he had entered a race, yet having seen and learnt alot. This would illustrate somewhat how I have come to this thesis. A lot of differentthings have kept me busy during the making of this thesis, sometimes losing sight ofwhat was happening around me, but learning a great many things. Therefore, to startof, I would like to thank everybody, that has aided me during the process, and some-times pointed me back in the right direction. It has been truly interesting, fun, and veryinformative.

I would like to thank Christian, of course, for his enthusiasm and helping me insuch a way, that I still had to think about what it was I had to do. I have walked out ofyour office many times, thinking I knew the answer, only to discover, it wasn’t quiteas straight forward as it seemed. I would like to thank you sincerely for making thisthesis an enjoyable experience. I would also like to thank the whole Superbus team,especially Antonia and Maarten, for the infrequent but highly enjoyable contact, andgiving me the opportunity to help you with the Superbus. The freedom and responsibil-ity you have given me in the making of this system, is greatly appreciated. FurthermoreI would like to thank Lennard for teaching me the practical skills of playing aroundwith expensive equipment. This was a lot of fun, and has been essential to turn theoryinto practise and the other way around. This gratitude is also directed towards Roel,who has given me a beautiful opportunity for a test campaign, and helping me a lot inthe process. Frank Boon from Septentrio is also thanked, in particular for the speedand remarkable openness of the responses from him and his company.

My sincere gratitude also goes out to all my friends and family, for helping me andalways being there. You have given me the confidence and perseverance to completethis study. This would not have been possible otherwise. I hesitate naming you allin person, in fear of forgetting or ranking anyone. Certainly my parents deserve anhonourable mentioning. They have given me opportunities that not many people have,for which I am extremely grateful. Also specifically for helping me the last few hecticmonths with proofreading and logistic tasks. My brother also, for sometimes carvingmy path, and sometimes the pointy but truthful remarks. A special thanks also toDennis, for helping me with the visual lay out of the report and presentation. Son, Ro,Behn, Nigel, Michiel, Matthijs, Klaas, Jelle, Geert, David, Daf, Joep, Barend, Noortje,Floor, Nathalie, Elisa, Claire, Doris, Lysanne, Vi and all the others, thanks for all thepep talks/proofreading/dinners/climbing and all the other forms of help and activities.It was great.

iii

Abstract

In this report a high accuracy positioning system is investigated for use in the Su-perbus. The Superbus project is an effort to apply a new and complete conceptualapproach to public transport. It consists of a vehicle, logistics and infrastructure. Thepositioning system for the vehicle is to have a horizontal position error, which will notexceed 5 cm in 95% of the obtained solutions. Secondary requirements included large(quasi-national) deployment area. Networked RTK positioning using GPS is shown tobe a valid means to adhere to these requirements. Real Time kinematic (RTK) posi-tioning allows for the required accuracy, while a base station network will allow forthe required deployment area using Pseudo Reference Stations (PRS). UMTS is shownto be potentially effective for the required wireless data transfer using the NTRIP pro-tocol. Testing confirmed adherence of the Superbus Positioning System to the re-quirements in a variety of real world scenarios. Furthermore, the receiver is shownto perform to manufacturer specifications. Difficult environmental conditions, such asurban areas and multipath are confirmed to have an effect on the position estimate incertain situations. It is however demonstrated that the receiver is capable of adheringto Superbus requirements in these situations, provided the initial ambiguity resolutionis correct. Temporary signal loss of all satellites (for example due to an underpass)is shown to inflict the need to reinitialise the ambiguity resolution algorithm, causinga temporary unavailability (around 20 seconds) of the precise RTK position estimate.Heading estimates are also established to be within specifications in good conditions.In high multipath conditions, or conditions with a low amount of satellites in view, theheading estimates exceed the manufacturing specifications with varying margins. Thisis possibly due to the fact that the ambiguities of the secondary antenna cannot be fixedin this scenario. Material testing showed that carbon fibre, the material that initiallywould cover the antennas in the vehicle, is highly unfit for this purpose.

v

PrefaceWriting a preface is a strangely personal affair. It is one of the last things one does,

before print. A lot of things that have occupied you during the making of the reportoften find a way in the text. For this reason, I have decided to write this part in Dutch.Some have (rightfully) commented that the text below is quite easily translatable inEnglish. Still, I have opted to keep it this way, since the language is closest to me, andthe majority of the people that will read this. I apologise to the people who are not ableto read the preface, but I hope that they will enjoy the rest of the thesis.

Tijdens mijn scriptie is mij vaak gevraagd, wat de Superbus nu precies is. Een vanmijn antwoorden was vaak, dat het een demonstratie was van de stand van de weten-schap. Wat is er mogelijk met wetenschap, en waar gaat het naartoe. Techniek, zoalsde Superbus, is bij uitstek een graadmeter voor de stand van de wetenschap omdat hetaantoont wat voor de mens beheersbaar is geworden. Dit beheersbare element is eensteunpilaar, een wetenschappelijke theorie moet falsifieerbaar zijn. Techniek is nietmogelijk zonder dat de uitkomst van een actie voorspelbaar is. Dit klinkt logisch, tochis wetenschap niet altijd verbonden geweest met falsificatie. Descartes heeft dit in1637 als eerste opgeschreven in zijn Discours de la Methode [8]. Dit wil niet zeggendat er daarvoor geen nuttige dingen zijn gezegd of uitgevonden, maar wel dat sinds-dien de kijk op de wereld en de wetenschap is veranderd. In de volgende paragrafenzal hier los op worden ingegaan [35].

In de vorige paragraaf zijn een aantal dingen impliciet opgenomen. Een daarvanis dat techniek klaarblijkelijk evolueert. Dit Darwinisme is een vreemd fenomeen,en het laatste -isme, dat door de moderne wetenschap wordt zonder hakken en stotenwordt geaccepteerd. De moderne samenleving ervan doordrenkt. Het kernpunt vanhet Darwinisme, van evolutie, is een gebroken eenheid. het hangt aan elkaar doorsterfelijkheid, vermenigvuldiging en verandering. Een nakomeling is een replicatievan zijn ouders. Deze mens niet hetzelfde als zijn ouders, maar men spreekt wel overhetzelfde, een mens. Mijn opa is gestorven, dat zal ik ook. Wij zijn niet hetzelfde,toch dragen wij dezelfde naam. Wij delen een identiteit, die groter is dan wij als in-dividuen. Ondanks dat een paard door de jaren heen is veranderd, blijft men sprekenvan een paard. Identiteit is veranderlijk, en wordt pas, juist door zijn vermenigvuldig-ing duidelijk. Darwinisme betekent een verandering van de kleinste gemene deler vanhetzelfde. Deze is met het Darwinisme omgewenteld van individu naar genoom.

Als gezegd viert Darwinisme hoogtij. Dit is omdat met de ontdekking van hetDarwinisme zeer veel wordt blootgelegd in taal, cultuur en techniek. het extendedphenotype van de mens [7]. Taal bestaat ook uit een gebroken eenheid. De betekeniswordt duidelijk door zijn vermenigvuldiging, tevens is de identiteit ervan evengoedveranderlijk. Een woord dat slechts eenmaal wordt gebruikt, is betekenisloos. Tevenskan een woord kan uitsterven, en een betekenis kan veranderen. Wij gaan er vaak va-nuit dat taal utilitaristisch is, dat wij controle hebben over de taal. Dit kun je niet meervolhouden wanneer je het bovenstaande accepteert. Voorbeeld, een kind voegt zich in,in de heersende taal. Deze is groter dan zichzelf, en hierdoor wordt het onmogelijk om

vi

”om het hoekje” van je eigen taal te kijken. Je bent al ingelijfd in de taal, voordat jeer over wilt praten. Dit illustreert dat techniek cultuur en taal met de mens leven, maardat de mens geen controle heeft over het verloop ervan. Dit ligt aan alle heersendeomstandigheden.

Is de mens hier niet de factor die besluit wat juist en onjuist is? Is ze niet de ratiovan techniek? Een gedachte experiment. Het betreft de vraag waar ratio en beteke-nis op komen zetten. Wat betekent iets, en daarmee, wat is iets. Het kan vrij simpelworden verwoord. Stel wij hebben twee simpele en gelijkende organismen A en B. Bwijkt op een cruciaal punt af van A, doordat B de mogelijkheid heeft een veranderingin zuurgraad te detecteren. Hierdoor heeft B de mogelijkheid om zich uit de voetente maken als het heersende milieu hem niet zint, waarbij A zich overlevert aan purekans. B vergroot hiermee zijn overlevingskans en zal ter zijner tijd zege vieren in hetgevecht om dezelfde, beperkte levensomgeving.

Het is belangrijk te zien dat door deze ontwikkeling een betekeniswereld wordtgeopend. Het is plotsklaps zinvol geworden om over zuur en niet zuur te praten. Hetbeınvloed je overlevingskansen. Daarvoor kon je niet spreken over zuur en niet zuur.Er was geen verschil tussen, het was betekenisloos. Met de mogelijkheid tot detectie,ontstaat zuurtegraad, plaats en tijd, juist doordat er consequenties aan zijn verbonden.

Hiermee wordt duidelijk dat een replicator (genoom) een ratio heeft boven die vanhet individu (replicant). Je kunt niet anders dan zeggen dat het genoom van soort B (pertoeval) rationeel is geweest. De ontwikkeling was het goede antwoord op de heersendeomstandigheden. Soort A heeft verloren, soort B leeft voort. Zo is de ontwikkelingvan de vleugel ook een juist antwoord op de heersende omstandigheden. Stukje bijbeetje heeft zich dat steeds verder ontwikkeld tot iets waarmee een vogel kan vliegen.De uitkomst had ook niet iets anders kunnen zijn dan iets wat lift genereert, want eenvleugel moet zich houden aan de wetten van de wereld om zich heen. Je kunt dusook niet anders zeggen, dan dat een vleugel er is om te vliegen. Het is onlosmakelijkverbonden aan zijn functie. De natuur heeft het bij het rechte eind.

Dit kan geextrapoleerd worden naar techniek. Een hamer (ook een replicator)heeft zich in de geschiedenis ontwikkeld en is immer bijgebleven met de heersendeomstandigheden. De botte steen is uitgestorven, en de hout met stalen constructie leeftvoort, daar ze het meest passend is voor de heersende omstandigheden. Het woordhamer blijft, de fysieke verschijning ervan verandert.

Ook voor wetenschap geldt hetzelfde. Teruggrijpend op de geschiedenis: Descarteskwam zoals eerder vermeld, met de wetenschappelijke methode. Een manier voor hetbedrijven van wetenschap. Falsifieerbaarheid kwam hiermee hoog in het vaandel. Ditduidde op een eerste radicale omwenteling van het wereldbeeld: de mechanisering(de tweede is het Darwinisme). het was een omwenteling in de identiteit van dingen.Het veranderde de blik van wat iets is en maakte het onafhankelijk van geloof (bi-jvoorbeeld de Ideeenleer van Plato met een niet fysieke wereld met de essentie vanalle dingen erin) en trok het naar reproduceerbaarheid. Water was niet meer per s eenstof met een essentie. Neen, water is een stof die gaat koken bij honderd graden. De

vii

eigenschap kun je reproduceren, en gebruiken. Maar belangrijker: Een tripje naar dealpen maakt het tegendeel duidelijk. Toch, de wet is niet ongeldig geworden dooreen tegenspraak. Ze is er tegen bestand. Ze is ingedekt om tegenstand te weerstaan,kan veranderen en variaties in zich opnemen, naar mate de wetenschap voort schrijdt.Een hypothese blijft bestaan als ze de beste resultaten oplevert. Iets is hiermee bij ze-gen van zijn reproduceerbaarheid en de potentiele gebruiken ervan. De wetten die ditbeschrijven zijn ook onderhevig aan het Darwinisme. Slechts de meest passende hy-potheses zullen overleven in de strijd van de wetenschap. herinnert u zich Phlogistonnog?

Descartes zei het onopgemerkt zelf al. Hij probeerde, levend in het tweespalttussen klassieke en wetenschappelijke wereld, nog een opening te houden voor klassiekeopvattingen, maar daar faalt hij in. Hij zag de wetenschap als boom [9]. Met een dikkestronk, en uitbreidend in steeds kleiner wordende takken. De stronk was voor hemde basiswetenschap, wiskunde en fysica. Van daaruit kon elke wetenschap en kennisworden afgeleid, de steeds maar kleiner worden takken. Hij doorzag echter al dat elkewetenschappelijke hypothese uiteindelijk wordt getoetst uit de techniek die het voortbrengt: de vruchten. Deze techniek dicht het gat tussen mens en natuur, met wie-len, hamers, telefoons en internet. Deze vruchten zijn succesvol. De vruchten voedendaarmee de boom voor de immer voortschrijdende wetenschap. Wederom Becher metzijn Phlogiston. Zijn simpele elementen systeem was op de lange duur niet afdoendemeer, de tak baarde op den duur minder vruchten dan concurrerende hypotheses, enstorf af. Ze is nutteloos geworden

Descartes, in zijn tweespalt, kon zijn boom echter niet in de lucht laten zweven, enzocht naarstig naar een (naar het blijkt onnodige) grond voor de wortels. Deze grondvond hij in de religie en de filosofie (metafysica). Hiermee voert hij helaas een totaaloverbodige dubbele boekhouding in. Zijn theologische en filosofische gronden voor dewetenschap zijn onnodig wanneer het succes van een wetenschap slechts beoordeeldwordt uit het resultaat.

Waar komt Superbus in dit verhaal? Dat kan alleen de toekomst vertellen. Deidentiteit van de superbus moet nog duidelijk worden. Duidelijk is wel dat het eenmooi voorbeeld is van het resultaat van de wetenschap. Bij uitstek een voorbeeld vanwaarom ik techniek ben gaan studeren. Een apparaat dat ogenschijnlijk de wetenschaptart. Over enkele jaren zullen we niet meer verbaasd zijn over dit project. Dan wekkenandere dingen onze interesse, oud nieuws. Dat wil niet zeggen dat de Superbus ver-geten is. Dat ligt eraan of Superbus een succesvolle replicator gaat worden.

Contents

Contents ix

List of Figures xi

1 Introduction 1

2 Superbus 32.1 The Superbus project . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Conceptual design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.3 Superbus and positioning . . . . . . . . . . . . . . . . . . . . . . . . 132.4 Superbus positioning requirements . . . . . . . . . . . . . . . . . . . 132.5 Superbus positioning project definition . . . . . . . . . . . . . . . . . 14

3 Theory Of High Accuracy GPS 173.1 GPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.2 Standard Positioning Service . . . . . . . . . . . . . . . . . . . . . . 223.3 Measurement error sources . . . . . . . . . . . . . . . . . . . . . . . 263.4 Augmented positioning . . . . . . . . . . . . . . . . . . . . . . . . . 283.5 The LAMBDA method and future enhancements . . . . . . . . . . . 393.6 Pseudorange rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.7 Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.8 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Superbus Positioning System 554.1 Introduction to the Superbus Positioning System . . . . . . . . . . . . 554.2 NTRIP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564.3 Superbus Positioning System . . . . . . . . . . . . . . . . . . . . . . 574.4 The Position Module . . . . . . . . . . . . . . . . . . . . . . . . . . 644.5 GPS applications and future developments . . . . . . . . . . . . . . . 71

5 Validation of Superbus Positioning System 795.1 Introduction to validation . . . . . . . . . . . . . . . . . . . . . . . . 795.2 Performance requirements . . . . . . . . . . . . . . . . . . . . . . . 845.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

ix

x CONTENTS

5.4 Test results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

6 Conclusion 135

7 Discussion and Recommendations 139

Bibliography 143

Acronyms 149

Appendices 153

A Septentrio PolaRx2e Specifications 155

B Septentrio PolaNt Specifications 159

C Antenna radiation pattern 161

D Example output Position Module 165

E Angular velocity precision estimation 169

List of Figures

2.1 Front and side view of a Superbus vehicle. [source: Superbus] . . . . . . 52.2 Comparison between a Superbus and a conventional bus. [source: Superbus] 102.3 Gull-wing door design. [source: Superbus] . . . . . . . . . . . . . . . . . 112.4 Ground clearance in high and low settings. [source: Superbus] . . . . . . 11

3.1 Segments of the GPS system. [source: NATIONAL ACADEMY PRESS[2]] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Graphical representation of a multipath situation. [source: BKG] . . . . . 273.3 Graphical representation of an RTK set up. . . . . . . . . . . . . . . . . . 333.4 Graphical representation of relative positioning errors. [Source: 06-GPS] 383.5 2D example constant cost ellipsoids before (a) and after (b) the decorre-

lation step. Rounding in the first scenario would produce incorrect inte-gers of (4,5), while rounding in the second scenario produces the correct(−2,10). [Source: [31]] . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.6 Cell structure of cellular networks. [source: [5]] . . . . . . . . . . . . . . 483.7 Representation of frequency shift modulation. [source: wikipedia] . . . . 48

4.1 Generic NTRIP set up. . . . . . . . . . . . . . . . . . . . . . . . . . . . 584.2 Superbus GPS system lay out with dedicated base station. . . . . . . . . . 604.3 Superbus GPS system lay out with network based VRS. . . . . . . . . . . 624.4 Septentrio PolaNt antenna and the PolaRx2eH receiver. [source: Septentrio] 634.5 Placement of the two GPS antennas in the Superbus. . . . . . . . . . . . 654.6 Position Module, high level lay out. . . . . . . . . . . . . . . . . . . . . 674.7 Vehicle behaviour during cornering. Active suspension is enabled on the

right. [source: Bose] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.8 Superbus FMS example. . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.1 Bias and precision graphically represented, accuracy = precision + bias.[source: wikipedia] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5.2 Outlier detection. [source: Septentrio] . . . . . . . . . . . . . . . . . . . 825.3 Location of the material test. [source: Google Earth] . . . . . . . . . . . 885.4 Number of satellites visible during materials test (cut-off angle 5). . . . . 88

xi

xii List of Figures

5.5 Rover antenna covered with carbon fibre material,supported by the woodenconstruction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

5.6 NMIbuilding and test set up. . . . . . . . . . . . . . . . . . . . . . . . . 915.7 Overview of the top of the NMI building. . . . . . . . . . . . . . . . . . 915.8 Vehicle frame overview, dimensions and antenna placement. Vehicle front

is top of figure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945.9 Vehicle with installed frame. . . . . . . . . . . . . . . . . . . . . . . . . 965.10 Skyplot over Delft, during the materials experiment. . . . . . . . . . . . . 985.11 Comparison of three, 30 minute, static solutions. . . . . . . . . . . . . . 995.12 Comparison of the dynamic solutions. . . . . . . . . . . . . . . . . . . . 1025.12 Comparison of the dynamic solutions. . . . . . . . . . . . . . . . . . . . 1035.13 C/N0 (signal strength) ratios of two satellites under three circumstances. . 1045.14 Skyplot over Delft, during the static experiment. . . . . . . . . . . . . . . 1055.15 X position and heading error over time. . . . . . . . . . . . . . . . . . . 1065.16 Speed error over time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1075.17 Skyplot over Delft, during the reacquisition experiment . . . . . . . . . . 1085.18 Demonstration of incorrect RTK ambiguity fix. . . . . . . . . . . . . . . 1085.19 Heading error due to unfixed secondary antenna. . . . . . . . . . . . . . 1105.20 Detail: Divergent heading error before loss of RTK. . . . . . . . . . . . . 1105.21 Histogram of the reacquisition times. A total of 56 trials . . . . . . . . . 1115.22 Cumulative representation of reacquisition times. . . . . . . . . . . . . . 1125.23 Track of the Duifpolder experiment. [source: Google earth] . . . . . . . . 1125.24 Skyplot over Delft, during the Duifpolder experiment. . . . . . . . . . . . 1135.25 Position error in relation to PVT Mode. . . . . . . . . . . . . . . . . . . 1185.26 Heading error in relation to PVTMode and NrSV. . . . . . . . . . . . . . 1205.27 Receiver speed comparison of the Duifpolder experiment. . . . . . . . . . 1215.28 Septentrio speed in relation to heading errors. . . . . . . . . . . . . . . . 1215.29 Track of the Emerald experiment. [source:Google Earth] . . . . . . . . . 1225.30 Skyplot over Delft, during the Emerald experiment. . . . . . . . . . . . . 1235.31 Septentrio error to ground truth in relation to the PVT Mode. . . . . . . . 1245.32 Septentrio loss of solution behaviour. . . . . . . . . . . . . . . . . . . . . 1255.33 Heading error in relation to the NrSV and PVT Mode. . . . . . . . . . . . 1265.34 Heading error in relation to the NrSV and Velocity. . . . . . . . . . . . . 1265.35 Heading error detail, showing dependence to the NrSV. . . . . . . . . . . 1275.36 Heading error detail, showing dependence to the NrSV. . . . . . . . . . . 1275.37 Track of the A4 experiment. A Viaduct is present at the bottom of the

track, an aqueduct crosses the halfway point of the track. [source: Googleearth] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5.38 Skyplot over Delft, during the A4 experiment. . . . . . . . . . . . . . . . 1305.39 Total Position error during A4 with respect to PVT Mode. . . . . . . . . . 1315.40 Cumulative time to first fix probabilities, absolute and RTK fixed. . . . . 1325.41 Heading errors versus PVT Mode and Number of satellites in view. . . . . 134

E.1 Angular velocities as experienced by the Superbus. . . . . . . . . . . . . 169

Chapter 1

Introduction

For several decades now, the major cities in the west of the Netherlands have beengrowing steadily in prosperity and population, while the north of the country has seena relatively slower growth. Especially in times of a beneficial economic climate thegrowth of the northern provinces has been lagging behind. In 1997 this was also theconclusion of a government commission that suggested that the profits, gained fromthe vast gas deposits of the north, could be used to improve infrastructure in theseareas. One of the incentives involves a better and faster transportation link betweenthe west and the north of Holland. By doing this, the hope is that the northern citieswill start to see larger economic growth figures, due to much faster and easier trafficbetween the north and the economic heart of the west.

The Superbus project started in 2004 as a reaction to this incentive, as well as toother modern day environmental and mobility issues: pollution, congestion and safety.The Superbus project sees to tackle these issues by applying a new and complete con-ceptual approach to public transport. It consists of a vehicle, but also new dedicatedinfrastructure and new logistics.

The Superbus itself is a high tech, road going vehicle that has been designed to befast, safe, comfortable, and flexible in order to promote its usability. Furthermore ithas been designed to have as little environmental impact as possible. It is to attain acruising speed of 250 km/h and will be powered electrically.

With the technology present in the vehicle, a demand also followed for a real time,high accuracy positioning system. Requirements were set according to which the hor-izontal position error of the system may not exceed 5 cm in 95% of the obtained so-lutions. This system will allow Superbus subsystems to function optimally and canallow for more advanced future upgrades, to the vehicle and the whole Superbus sys-tem. The purpose of this report is to investigate a positioning system, that can adhereto the requirements set by Superbus. Additionally the designed system will be testedto validate it’s suitability for use in the vehicle.

The structural composition of the report is as follows. Chapter 2 will serve as anintroduction to the Superbus, and the project as a whole. System requirements will

1

2 CHAPTER 1. INTRODUCTION

be specified, and the project goals will be defined precisely. Chapter 3 will constitutea theoretical background to the project, and will explain what systems are necessary.Chapter 4 will describe the positioning system and its practicalities. Chapter 5 willserve as the main chapter wherein the obtained system is tested and validated. The lasttwo chapters will consist of a conclusion, discussion and of recommendations.

Chapter 2

Superbus

Public transport plays an important role in modern society, it has become a backbonefor both social and economic interaction. This infusion in society does however notmean that changes and improvements in public transport are not possible or desirable.This is the vision of the Superbus project, it aims to re-establish the way we look atpublic transport.

This chapter will function as an introduction to the Superbus project. The origin ofthe project will be considered shortly, and Superbus concept will be introduced. In thesubsequent sections an introduction to Superbus positioning will be made, after whichthe requirements and positioning project will be defined.

2.1 The Superbus project

The Superbus project started in 2004 as a reaction to modern day environmental andmobility issues: pollution, congestion, time constraints and safety. These are issuesthat gained tremendously in importance in the last decade. The project will try totackle these mobility issues from more than one perspective. The project really tookoff in November of 2005 when it received a grant from the Ministry of Transport, Pub-lic works and Water management (Ministerie van Verkeer en Waterstaat), currently theMinistry of Infrastructure and Environment. The grant was conceived for the devel-opment of a means of public transport between the north of the Netherlands and themore densely populated south-west. This means of transport should function as aneconomic and social catalyst to draw the historically more isolated northern part of theNetherlands and the south-west closer to each other. Superbus was one such means oftransport selected that could fulfil this role.

After the grant was received, the detailed development got underway and in the be-ginning of 2007 construction began on a first test vehicle. Development of the detaileddesign continued throughout the building process, which is now nearly complete. Af-ter its completion the vehicle will undergo rigorous testing to analyse performanceand ensure passenger safety. During this time Superbus will try to develop enough

3

4 CHAPTER 2. SUPERBUS

knowledge and momentum for the concept to make implementation of the concept apossibility.

The Superbus project is a project that tries to rethink public transport. The goal ofthe project to provide a fast, demand based road transportation system for medium tolonger distances up to approximately 250 kilometres. It tries to address modern mo-bility demand with a fast, safe and environmentally friendly public transport vehicle.The vehicle is specifically designed with speed, passenger comfort and sustainabilityin mind. It is important to note that it is a complete conceptual approach to publictransport, and as such the project does not constitute a transport vehicle alone. TheSuperbus project spreads over multiple disciplines:

• Vehicle

• Infrastructure

• Logistics

• Safety and Reliability

• Environmental aspects

• Exploitation and economic viability

In order to discuss the concept of Superbus, the background is enlightened first.The demand for transportation grows constantly, as does the amount of people whodemand it. People have been travelling further and further over the years, caused bytechnological advances. Yet for daily commutes, time is a more important factor thandistance. People generally do not wish to spend more than a certain amount of timeeach day on transportation, or travelling [29]. This situation is now aggravated by anever growing fleet of cars dressing the nation, causing bottlenecks and traffic jams inthe established road network. Conventional growing of this infrastructure, buildingmore roads for more cars, will not be a viable, sustainable solution to the problem.Not only due to space constraints, but also due to safety issues and pollution. Super-bus therefore tries to find one of the solutions to this problem. It does so by studyingthe reasons for people to take cars and tries to derive an feasible alternative. At thesame time the solution must be sustainable, both for people and the environment.

An important element is to find the appealing and deterring factors of both privateand public transport. For example: the sense of privacy one has in a private vehicle, aswell as the fact that one is able to drive right to the door of your destination, are com-pelling factors for a automobile. On the other hand, a person driving is not able to dowork, for example. Public transport can in some cases allow for this, as well as avoidmany known traffic bottlenecks. If a new form of public transport is to be successful itwill need to outweigh the advantages of a car and/or negate the annoyances of publictransport.

It is clear that the problem has multiple facets, there is no single source that causesthe problem. As mentioned, the Superbus project sees the solution in several facets as

2.1. THE SUPERBUS PROJECT 5

well. For example: moderating the overall road congestion by diminishing the numberof vehicles on the road, increasing comfort and adding speed. The aspects mentionedin section 2.1, will all be combined to a system that is designed to provide fast, easy,comfortable and sustainable mobility to the user.

Very concisely: Superbus consists of a vehicle together with infrastructure and isdesigned as a more flexible yet fast alternative (mainly) to modern high speed trans-port. It aims to be more flexible by applying an on-demand structure, in contrast to, ortogether with a normal time-table. Starting points and destinations are therefore moreflexible, and can be more local due to the ability of the vehicle to use normal roads inaddition to its own infrastructure.

The Superbus (figure 2.1) itself is an electrical vehicle which can transport up to23 passengers at a speed of up to 250 (Km/h). The vehicle will be powered by bat-teries and will make use of separate infrastructure to reach these high velocities, theSupertrack. Thanks to the low weight, low aerodynamic drag and rolling resistancethe vehicle is very energy efficient. The absence of exhaust fumes and the low energyuse also make the Superbus an environmental friendly way of transport.

Figure 2.1: Front and side view of a Superbus vehicle. [source: Superbus]

Passengers are able to book a fare by means of internet or telephone. A centralbooking system combines passengers for matching destination and departure pointsand offers several travel options. The passenger can then book the most suitable, andhas a journey with little stops or transfers. Passengers are boarded on the Superbus inurban areas on normal public roads before the vehicle goes to the Supertrack where ittravels at high speed until it almost reaches destination. Here it transfers to local roads


to let the passengers of close to their end destinations.

This condensed section is of course not the complete report of the Superbus system.In the section below an overview of the concept will be sketched, clarifying problems,visions and design choices in no particular order. The sections will hopefully create aninsight in, and understanding of the main design philosophies of the Superbus system.

2.2 Conceptual design

A basic view of the Superbus concept has been presented in the previous section.This shows only the final result, provided the project is successful. In the followingsections, a more elaborate view will be laid down in no particular order. It can create aninsight into the design decisions and increase the overall understanding of the concept.Information for this section is redacted from Superbus documents [26–29, 52].

Travel times

First of all Superbus will try to reduce travel times in two ways. Not only does itincrease the speed of motion compared to normal cars and trains, but it also tries to de-crease transit,waiting, and initial transportation times (door-to-station). These can addup significantly in conventional public transport. The last method of reducing traveltime is the use of dedicated Superbus infrastructure, Supertracks. These tracks allowthe vehicle to reach its top cruising speeds, and circumvents normal traffic. In short,Superbus tries to reduce the door to door travelling times, not only the station to stationtimes.

With a cruising speed of 250 (km/h), Superbus will be competitive with most highspeed train services. The travel times are additionally reduced by altering the con-ventional time tables and itineraries of classic public transport systems. These mainlyemploy fixed line services with predetermined stops. Superbus will employ an on-demand structure. This means that they are not bound to fixed time tables. Becausethe vehicle is not limited to tracks as are high speed trains, this means that they arealso means that the fixed itinerary and destination becomes obsolete. The Superbuswill potentially be a much more flexible way of travel.

The on-demand structure can be exploited to benefit the user. The user will be ableto convey base and destination points to the Superbus system, as well as departure-and or arrival times. The Superbus system is then able to pool multiple users togetherand devise a optimised itinerary which will minimise travel times. It will also avoidtransits to a great extent, or optimise them, as to minimise the discomfort. The viabil-ity of such an on-demand system is aided by the fact the Superbus vehicle is able tocarry a relatively low number of passengers for a public transportation vehicle: 20 to30 depending on the lay-out. This allows customisation of routes and travelling times,while still reducing the number of vehicles on the road significantly: A four personpassenger car is on average only occupied by 1.4 people, a much smaller number thana Superbus [28].

2.2. CONCEPTUAL DESIGN 7

Logistics

Note that the Superbus will not be a pure door-to-door transportation vehicle. Collect-ing each passenger individually would take too much time and negate the time won bythe speed of the vehicle. Yet by being fully road-going, stopping points can be muchmore dynamic than for example trains. These points are called concentration points.These can be created by introducing fixed stations, but they can always be dynamicas well: when- and wherever enough people are present to justify a stop, one can becreated. Examples include events and conventions. Furthermore, because the Super-bus can only carry 20 to 30 people (depending on configuration), these concentrationpoints can be relatively small and local.

When the Superbus system is fully deployed and operational, it will employ a di-rect point to point routing system, with potentially a few local stops at the startingpoint and the destination. This will ensure fast travel times while maintaining a highdegree of vehicle occupation and minimising the distance a user needs to travel to getonto the vehicle.

This on-demand system does however not mean that standard line services willbe ignored completely. If demand is stable and large enough, such a line service canbe employed with little drawbacks, or even benefits, to the passenger. This can beof particular interest to commuters seeking for a stable fast way of travel for mediumto long distances. An example here, might be the Zuiderzeelijn, the original routeplanned for the Superbus, mentioned in section 2.1. Also in the beginning roll outphase of the system, not enough vehicles will be available to employ an effective on-demand system. Additionally, the passenger base needs to have grown sufficientlylarge to support the system. This is a two way system that will need time to grow.

One can see that if the concept takes off, the success will largely be dependent onmodern ICT solutions. Managing many passengers individually, providing each withan efficient travel solution, but also managing an increasing Superbus fleet, includingdestination, route and current location, will need a new, highly optimised, highly con-verged, and continually up-to-date system. The old time table system is thrown out,yet it does need to be replaced with something superior to benefit the passenger. Thisin order to become a viable alternative to other forms of transport.

Lastly the charter market may become a lucrative market for Superbus. Since thevehicle has a relatively low number of passengers, it becomes more obtainable fora single business to acquire a high occupancy rate (or load factor). At this point itmay become beneficial to charter the complete vehicle. This will allow more freedomto the business and its users to travel on desired times and to desired destinations.This will again shorten travel times since the Superbus can use a direct route to thedestination. This may especially be interesting to medium to large corporations withseveral branches within the range of the Superbus.


Passenger comfort and safety

Benefits over other modes of transport, as already discussed in previous sections, areparamount for the success of the Superbus system. For example, the locality, and indi-viduality of the Superbus approach mentioned above are important steps in the passen-ger’s perceived comfort, and likely diminish the reluctance to opt for public transport.The Superbus concept tries to do this in other areas as well.

The vehicle itself will be extremely comfortable and is produced to express a cer-tain luxury. It must offer a significant benefit over a car in order to convince peopleto take the Superbus. Therefore the seats are very comfortable and head- and legroomare very spacious. However, the sense of privacy in a car is not to be underestimated.The seats are spaced far apart and many individual entrances drape the vehicle forthis reason. It will relieve some of the annoyances people perceive when using publictransportation. People must not feel hindered while travelling, be it for privacy, com-fort or the ability to work.

The ride itself must also not be overlooked. A harsh journey can issue the samefeeling and discomfort, and makes any form of productivity impossible for the passen-ger during the trip. First of all, the suspension design is a major factor for perceivedcomfort. Seconly, accelerations in all directions need to be managed: Longitudinal,lateral and vertical. This is shown in table 2.1.

Table 2.1: Recommended acceleration domain. [Source: Superbus [28]]

Direction domain (m/s2)

Longitudinal ± 1.0Lateral ± 1.0vertical 9.31 - 10.81

After analysis of comparable modes of transport, Superbus established that accel-erations need to remain in this domain in order to be perceived as comfortable. Thereare several ways to achieve this. First and foremost, de- and acceleration should bekept within these limits during normal operation. Superbus will not be a race car: itwill take a leisurely 70 seconds to accelerate from stand still to the cruising speed of250 (km/h) and vice versa. The vehicle is able to perform far better during emergencysituations, but naturally these situations tend to be avoided.

Lateral accelerations are harder to control, they are dependent on the speed of thevehicle and the radius of the turn, see equation 2.1. With a cruising speed of almost 70(m/s) this would leave a turn radius of 4.8 kilometres.

a =V 2

R(2.1)

The turning radius or speed can may be altered, if necessary, by banking the road.This will convert some of the lateral accelerations to vertical accelerations. This will


allow tighter turns at greater speeds. Note that banking roads would on the other handadd to the cost of the infrastructure. Since one of the main advantages of the Superbusconcept is the relatively low infrastructural investment, this may need to be avoided.

Another approach to this can be taken as well. In addition to the banking roads,the same results can be achieved by banking the vehicle: the same transition of forcesapplies. An active suspension system is able to raise, lower, soften and stiffen the sus-pension of each individual wheel. This would allow the vehicle to bank into the turnby raising the suspension on the outer wheels slightly. This will reduce the bank angle,and hence reduce the cost of the road. Note that this scenario is supposing the tractionof the wheels is sufficient, and the aerodynamics are not disturbed beyond acceptablelimits.

The active suspension (atlhough currently not implemented) is useful for control-ling vertical accelerations as well. Whenever a known coarseness in the road is en-countered the suspension of the appropriate wheels can be softened moments prior, toallow the vehicle to coast over the unevenness due to the low weight of the wheel. Thiswould significantly reduce the vertical accelerations, allowing the vehicle to remain athigher speeds while passing these variations in the road. This in turn would again re-duce the cost of the infrastructure since the need for a perfectly even road is no longerthere.

All the measures mentioned above combined try to relieve the reluctance to takepublic transport and offer genuine advantages of choosing the Superbus over a normalcar. But this comfort means very little without safety and reliability. Therefore safetyis a priority both in infrastructure and in the vehicle itself.

The fast Supertracks, the independent roads only accessible for Superbus, will bemonitored actively. Cameras will control the entrances and fences or noise barrierswill make access for wildlife and people difficult.

The vehicle too will have many active safety features embedded. Computer sys-tems will monitor more than 750 sensors, for example in the doors and seat belts ofevery passenger. These will affect the passengers directly. But all other on board partsand systems will be checked continuously as well. Tire pressure, engine and batterystatus, etc. Also seven radars will be fitted onto the Superbus to improve situationalawareness. At 250 (km/h) it will take a significant length of road to stop, some 2.5kilometres. The radars will aid in detecting obstacles and irregularities before the pilotis able to, and can aid in taking appropriate action. Passive safety is also not forgotten.Mandatory seat belts were mentioned already but also the full carbon fibre body will,if necessary, protect the passengers.

Sustainability

Conservation of the environment has become a relevant issue in recent years, and as aresult Superbus tries be as close to a environmentally neutral solution as possible. Thiswill ideally serve as an incentive for others, and show that an environmentally viable


solution can also be an economically viable solution. Naturally, the low environmentalimpact will increase the marketability of the Superbus as well. Yet most importantly,it will ensure the Superbus will remain a viable option in the future, as legislation oncarbon emissions and general environmental impact will invariably tighten.

The vehicle itself is electrically powered and offers regenerative braking, in orderto regain some of the kinetic energy of the vehicle. Electrical motors can convert en-ergy to movement with greater efficiencies than conventional motors [28]. Also, theelectrical energy can come from any number of (environmentally friendly) sources,wind, sun, etc. It is therefore less dependent on liquid fuel prices alone, in additionto being more efficient. Lastly, the aerodynamics are improved dramatically to reducedrag, and hence the power needed to attain the higher cruising speeds. The vehiclewill approximately need the same power output at 250 km/h as a normal bus needs at100 km/h [28].

To achieve the lower power requirement at speed, the frontal area is reduced incomparison with a conventional bus, and the aerodynamic properties are optimisedto reduce the drag coefficient. This is can be seen in figure 2.2. By using advancedmodelling and simulation software the drag coefficient has been reduced to one that islower than a typical passenger car [28].

Figure 2.2: Comparison between a Superbus and a conventional bus. [source: Super-bus]

The vehicle

To allow for the high speed performance of the Superbus while still managing a rea-sonable power requirement, a low, light weight and aerodynamically efficient designis mandatory. This has consequences for the interaction with the vehicle.

As can be seen in figure 2.2, the vehicle is obviously too low for a person to standin. This invalidates a vehicle set up with one set of doors and an aisle. Instead eachrow will have two individual gull-wing doors, allowing persons up to a length of 2.10metres to enter the vehicle in a normal manner and sit down naturally without havingto bend down. This can be seen in figure 2.3. It allows you to stand while the doors areopen, yet are low and flush when closed. The multiple door design however does meana more difficult design in order to maintain an ample structural strength and stiffnessfor the vehicle. The doors are formed hexagonally to optimise the transition of forcesin the vehicle frame. In order to incorporate this design without compromising in theweight restrictions, the vehicle will have a fully load bearing carbon fibre construction.


The hexagonal structure allows for this, and obtains sufficient torsional and longitudi-nal stiffness.

Figure 2.3: Gull-wing door design. [source: Superbus]

A problem arising from the extremely low vehicle design is a that a reduced num-ber of roads is accessible. In order to make Superbus useful, it should at least be ableto travel on all roads a normal bus is able to go. With the chassis just mere centimetresfrom the ground this is not possible, Speed bumps, steeper bridges, and other urban ob-stacles would be impossible to navigate. Therefore the complete vehicle can be liftedup or down nearly 40 centimetres. Low for high speed cruising, high for inner cityflexibility, see figure 2.4.

Figure 2.4: Ground clearance in high and low settings. [source: Superbus]

Now that most external design choices are clarified somewhat, it may be usefulto summarise some of the vehicle specifications and performance parameters. This isdone in the table below.


Table 2.2: Superbus specifications overview. [source: Superbus]

Main specifications

Drive system 4 electric motorsPower Output 300 kW (600 kW max)Range > 250 kmAcceleration (0-100 km/h) 36 sMaximum cruising speed 250 km/hLength 15 mWidth 2.5 mHeight 1.65 mWeight 10000 kg incl. payloadSeating capacity 23

2.3. SUPERBUS AND POSITIONING 13

2.3 Superbus and positioning

Now that the Superbus concept has become clear, it is the appropriate time to discussthe goals of the positioning subsystem in the Superbus. Why is positioning needed inthe Superbus and what are the main tasks of such a positioning system?

Simply put, the primary task of the positioning system is to aid additional Super-bus (sub)systems. It will provide essential information to other Superbus systems andsoftware. In section 2.1 it is explained that the Superbus is more than a straightforwardvehicle, and encompasses many systems and sensors to ensure a comfortable and safejourney. Some of these systems need position, heading or other situational awarenessinformation in order to function. As an example one can think of something as simpleas navigation software, but the data can also used for more elaborate systems. Exam-ples will be given below. Even more information can be found in section 4.5

Position data will also become an integral part of a planned elaborate database sys-tem, incorporating various kinds of information concerning road conditions. This can,amongst others, result in a more effective driving strategies for the Superbus. Sucha database can for a example incorporate permanent road features, such as corneringstrategies and locations of road imperfections. But in later stages variable conditionscan be incorporated as well. Think of features such as current road temperatures, localweather conditions, and temporary road works or obstacles. All these features mayprove useful in a vehicle that promotes fast, safe and smooth transportation.

Other uses may be found in the efficient fleet management system required to sus-tain an on demand itinerary structure. Section 2.2 shows that this system could becomequite complex, resulting in the need to have up to date position information of all thevehicles. Lastly, as was mentioned in section 2.2, the vehicle is equipped with radarsthat provide additional situational awareness around the vehicle. Positioning informa-tion, more specifically, yaw rates, can be used for processing the raw radar data intointelligible information.

So in conclusion, the goal of this project is to provide precise heading, pitch, posi-tion, velocity and time information to the Superbus vehicle in order to support diversesubsystems. In addition supplemental information can be provided regarding the per-formance of the computed solution. These can aid in the safety issues concerning sucha fast moving vehicle.

2.4 Superbus positioning requirements

Throughout this chapter it has become clear that a positioning system is a necessityfor Superbus. In the previous section the essential physical quantities needed for theSuperbus subsystems have become more explicit. Yet the performance for these quan-tities is still unclear: What is the needed precision for these quantities? And althoughthese performance requirements of the Superbus GPS system are clarified in more


detail in section 5.2, it is useful to discuss the global requirements the Superbus posi-tioning system. It will provide a background to the next chapters, in which both thepositioning system as the theory will be explained.

Together, the Mathematical Geodesy and Positioning (MGP) section and the Su-perbus team, set up the general requirements, discussing what information was neededfor the Superbus systems and subsystems. Flexibility and precision were paramount,allowing direct use of the required data, but maintaining overhead, especially in preci-sion, for future developments. This resulted in the following requirement:

The Superbus Positioning System, must be able to provide real time,precise positioning data. Horizontal position error may not exceed morethan 5 centimetres in 95% of the obtained solutions.

One notices that this requirement only mentions positioning accuracies, for theproject however, speed, heading and pitch parameters with matching precisions wereupheld. Additionally the dynamic nature of the project, an update rate of 10 (Hz) wasupheld. As a last addition to these requirements, the deployment of the Superbus wastaken into account. Superbus is a fast moving vehicle, for trans regional transporta-tion. This means a large deployment area is an inherent character of the Superbus. Inaddition the Superbus system aims to be flexible in terms of destination, starting pointand routing. This has led the project to the adoption of a (quasi)national deploymentarea for the Superbus. In other words, the requirement above, should be valid in largeareas of the Netherlands.

As already mentioned, a more elaborate complete overview of the precise perfor-mance parameters is given in section 5.2. A more elaborate view on some subsystemsthat utilise positioning information will be given in section 4.5.

2.5 Superbus positioning project definition

Most important parts involving Superbus have now been clarified. The concept is clear,as are the uses and requirements for a Superbus positioning system on the global level.The positioning project itself however, still needs to be clearly defined. This sectionwill outline the exact goals, boundary conditions and prerequisites for this project.This section shows what needs to be done, and in the process provides an overview ofthe next four chapters.

Firstly, the boundary conditions of the project are discussed. In the initial phasesof the Superbus project it was decided that a practical and flexible approach to thepositioning system was to be taken. This means that from a very early stage it wasclear that a flexible Commercial-Of-The-Shelf (COTS) positioning solution was to beimplemented. Choosing this path ensures support for the system, while keeping de-velopment costs to a minimum. Together with the MGP section it was decided thatsatellite navigation would be the most suitable system for the positioning demands ofthe Superbus. This system, more precisely the Global Positioning System (GPS), isalready in use and could potentially adhere to the requirements of section 2.4. The

2.5. SUPERBUS POSITIONING PROJECT DEFINITION 15

satellite system is already available, and commercial equipment is readily availableand supported. Another advantage is that GPS is globally available. For Superbusthis means that this solution is independent on infrastructure and allows for (globally)flexible trajectories. The global nature of GPS can also be an advantage because thesystem needs to be certified only once, which is not the case for solutions dependenton the individual driving infrastructure, such as optical systems. This again makes theinfrastructure cheaper and easier to obtain, not every individual Supertrack needs to becertified individually. This on the other hand also means that the proposed solution isalready bounded to GPS in order to acquire a positioning information that meets therequirements. It directly places other potential solutions such as optical and inertialbased positioning in second row.

Now that it is clear that the project is bounded to GPS equipment, the positioningproject can be easily and sharply defined. The GPS dependency immediately sets cer-tain circumstantial boundary conditions on the system design itself.

Literature suggests that a stand alone GPS system is theoretically not able to attainthe accuracies required for the Superbus, using exclusively the currently available sig-nals [31]. For these accuracies, additional information, available from varied sourcesis mandatory. Therefore it was known that some form of data communication betweenvehicle and main land was mandatory.

A second circumstantial boundary condition is the following. The Superbus re-quirements include heading, pitch and yaw rates. These are unavailable from GPSsystem with one antenna, at least not directly. In order to make these available, a sec-ondary GPS antenna will always be a requirement.

The last boundary condition, is that the physical GPS receivers and antennas werealready determined by the time the author joined the project. This predominantly af-fects antenna placement in the vehicle and the interface with other Superbus hardware.

The requirements, conditions and hardware combined broadly outline the designdirection and determine the outline of the GPS project. This can be broadly subdividedinto four sections.

Firstly the physical part: System design and physical integration. This entails de-termining what the capabilities and requirements of the chosen receivers are, and inwhat way the required information is accessible. This, together with the next section,will in its turn help to determine the method of positioning used, and how the completesystem will fit together and may be implemented. But this will also determine the waythe receivers will output data and interface with other hardware. It will result in asystem available for physical implementation in the vehicle. The physical section alsoencompasses antenna placement, which can be crucial for the availability and accuracyof the calculated GPS solution. The placement therefore can play an important role inthe adherence to the requirements. Lastly this section involves selection of proper frontend and back end settings for the receiver. This entails setting up the receiver to ob-tain the best solution in the highly kinematic environment it resides. This is the back


end. But establishing which information to input and output, as well as determiningthe output rate, and output format is also important. This is called the front end.

The second part is data communication link. As mentioned, developing a GPS sys-tem that is theoretically able to abide the stated requirements, always needs additionaloutside information. The second section of the GPS project involves the selection andimplementation of this mandatory communications system. The Superbus propertiessuch as deployment area and speed, but also the needed data link bandwidth are im-portant factors in this field.

By this time, the Superbus GPS system will be physically largely determined,whereafter the focus shifts to software. The third section is therefore: software andinterface. This section implicates the interface with other Superbus hardware.

In the beginning of the project it was determined that a sample C++ program was tobe developed that was able to set up the receiver, as well as to receive and interpret thereceiver output. The project requires some information that is not always available inthe standardised output format: NMEA. The binary format developed by the receiver’smanufacturer, does give access to this information. This does however complicate theway the information is accessed, hence the sample C++ code.

The last section is the verification and validation of the completed GPS system.Establishing the performance of the system and juxtapose the measurements to boththe Superbus requirements and the manufacturer specifications in a series of tests.

The project sections described above loosely follow suit with the following chap-ters. The first and second sections will be discussed in the next chapter. After this, alltheory will be available. It is then applied and the complete system will be discussedin chapter 4. This chapter will also discuss the third section, the software. Lastly thevalidation and verification are done in chapter 5.

Chapter 3

Theory Of High Accuracy GPS

A Global Navigation Satellite System (GNSS) is a modern navigation aid that is ableto provide the Superbus with position, speed and heading data. Currently there are sev-eral such satellite systems in existence, in various stages of development. The GlobalPositioning System (GPS) however, is currently the only fully functional and globallyavailable system. It is developed by the American Department of Defence (DoD), andbecame fully operational in April 1995.

In this chapter, the GPS system will be explained. It will become clear that theSuperbus requirements, discussed in the previous chapter, are theoretically attainableusing augmented positioning.

The first sections introduce the GPS system and provide a theoretical backgroundon the Standard Positioning System that GPS employs. Subsequently the main errorsources in GPS are explained, after which we will discuss augmented positioning as away to adhere to Superbus requirements. Next, the pseudorange rates will be touchedupon briefly, whereafter the communication system needed for augmented positioningwill be discussed. A brief conclusion will finish the chapter.

3.1 GPS

GPS segments

The complete system architecture for the GPS system consists of three segments: Aspace segment, a control segment and a user segment, as can be seen in figure 3.1.

The space segment consists of a nominal figure of 24 satellites to provide globalcoverage, but currently constitutes 32 satellites (January, 2010). The satellites have anear circular, orbital radius of 26,560 (km) in 6 planes, with a 55 inclination, allowinga user with a clear view of the sky to always receive the minimum of 4 satellites. Eachsatellite has an orbital period of 718 minutes and travels at nearly 3.9 Km/s. The satel-lites are equipped with highly accurate atomic clocks, and every satellite continuouslytransmits radio signals that allow a receiver to measure and compute the distance to

17

18 CHAPTER 3. THEORY OF HIGH ACCURACY GPS

Figure 3.1: Segments of the GPS system. [source: NATIONAL ACADEMY PRESS[2]]

the satellite.

The control segment is a network of tracking stations, located around the globe,with the master control station located in Colorado, USA. The main function of thecontrol segment is to monitor and control the space segment satellites. The controlsegment predicts the individual satellite orbits and the behaviour of the atomic clocks(which vary over time) and updates the broadcasted navigation message accordingly.

The user segment comprises the GPS receivers, available for both military andcivilian users. These receivers are able to obtain a Position, Velocity, Time (PVT)solution by measuring the range (and range rate) to a minimum of four satellites inview.

Satellite signals

There are several possibilities to compute a receiver position by using GPS. All ofthese possibilities however, are dependent on the radio frequency (RF) signal origi-nating from the satellites. For this reason the architecture of the GPS signal will beclarified.

GPS satellites currently transmit signals over two frequencies, L1 and L2. Theindividual properties of both signals are listed below In table 3.1.

An L1 or L2 signal leaving the satellite is a combination of three respectively fourcomponents: a carrier wave, a Coarse Acquisition code (C/A code), a precision code(P(Y) code) and finally a navigation message. L1 consist of all four, while L2 lacksthe public C/A code (although new satellites are already launched with similar func-

3.1. GPS 19

Table 3.1: Properties of transmitted satellite signals

L1 signal L2 signal

Carrier Frequency (MHz) 1575.42 1227.60Wavelength (m) 0.19029 0.24421

Code Frequency (Mcps) 1.023 & 10.23 10.23PRN codes C/A & P(Y) P(Y)

tionality on the L2 band). These components will now be explained below.

Carrier wave

The carrier wave is the actual sinusoidal RF signal on which the other two componentsare modulated. It is transmitted at 1575.42 (MHz) for the L1 signal. L2 uses a 1227.60(MHz) signal, amounting to wavelengths of 19.03 to 24.42 centimetres respectively.

Pseudo Random Noise codes

The ranging code, or the C/A and P(Y) code, is a mathematical pseudorandom binarysequence. This is a sequence which is repetitive and highly orthogonal. Orthogonalityof a radio signal implies that a receiver is able to reject an arbitrarily strong signalwhen not coded in same coding scheme. Pseudo Random Noise (PRN) codes are onesuch coding scheme. The use of these PRN codes in the GPS system is significant fortwo main reasons. First, it allows all GPS satellites to broadcast over the same fre-quency. Each satellite has a separate unique PRN code. One such code is orthogonalto all the remaining PRN codes of the satellite system. This allows identification of thesatellite by the code. It also eliminates interferences between satellite signals, allowingall satellites to broadcast on the same frequency.

In case of GPS, orthogonality of the signal is exploited by recreating the exactsame code within a receiver. When two different PRN codes are multiplied, the resultwill always be around zero: the two signals will cancel each other. The only exceptionto this is when two identical codes are multiplied. When the two signals are correlateda large output will be the result. This is of great importance, and the second advantageto using PRN codes. The repetitive nature of the code and its correlation features canbe exploited. As mentioned, PRN codes can be replicated in an GPS receiver. Thisreplicated code is subsequently be shifted in the time domain (In reality they are alsoshifted in the frequency domain because of Doppler effects, this is however omittedfor now), until it lines up with the received signal. This enables the receiver to acquirethe signal travel time, and hence the range to the satellite. Note also that the satel-lite signals received by the receiver on earth are below the natural static noise levels,and are therefore essentially ”invisible” for radio receivers. However by shifting thereplicated signals in time and multiplying these with the received signals, the satellitesignal are amplified above noise levels, when -and due to the PRN codes only when-


the two signals are exactly aligned.

As stated above, there are two types of PRN code used in the GPS system. TheCoarse Acquisition code and the precision code. These two codes, have different char-acteristics and allow for different positioning precisions, however the same principleapplies to both. The C/A code is a publicly accessible code, allowing users all over theworld to use GPS signals for positioning purposes, the Standard Positioning Service(SPS). The precision code allows for more accurate positioning, however, this signalis encrypted to the Y code, and accessible to the United States military only.

The C/A code is 1023 bits long. A bit is also often called a chip. The code isrepeated every millisecond. This means each chip is approximately 1 microsecond, or300 metres long. The encrypted P code however is extremely long, about 1014 chips.The code is also sent at a faster pace, 10 times that of the C/A code, resulting in achip length of about 30 metres. Modern receivers are able to synchronise the satellitecode and the replicated code to approximately 1% of the length of a chip (0.1% formodern high end receivers). This gives an indication of the theoretical precision of thepositioning solution, for the moment disregarding any other errors, such as ionosphericdelays and multipath. For SPS this amounts to an approximate precision of about 3metres, while the precision service will be around 30 centimetres, due to the muchshorter chip length.

Navigation message

Finally, navigation data is present in the satellite signal. This is a series of messages,binary coded, modulated in the signal at 50 (bits/s). These messages contain the TimeOf Week (TOW), GPS week number, satellite health information, clock bias parame-ters, ephemeris data and an almanac. Ephemeris data, constitutes the satellite’s preciseorbit information. The Almanac contains coarse ephemeris data of all the satellites,and an ionospheric model allowing receivers to correct for signal delays encounteredwhen the satellite signal travels through the ionosphere. Because of the low data-rate ofthe navigation message, it takes 12.5 minutes to download all the data. The ephemerisdata, needed for receiver positioning can take as much as 30 seconds to load.

Observables

In the previous section the signal side of the GPS system is examined, but it is not yetmade clear how the receiver is able to use these signals. The PRN code and its cor-relation behaviour have been discussed briefly, the methods of positioning employedby receivers will become clear in the following sections. Here the most basic of re-ceiver results will be introduced: the measurements, or observables. These are thebasic building blocks for obtaining a PVT solution.

For now we are aware of two things: First of all, it was stated that observationsare necessary to ascertain any knowledge of your position. Secondly we now know

3.1. GPS 21

from the previous section, that GPS satellites broadcast modulated radio signals on 2frequencies.

It is also clear from the previous section that information is present within thetracked satellite signals. But what kind of data exactly can be extracted from a trackedsatellite? This data is divided into two groups:

• Pseudorange measurements, ρ

• Carrier phase measurements, φ

These parameters can be extracted for every individual tracked satellite, every sin-gle epoch, or unit of measurement. This information may then be used to calculatea positioning solution, for that given epoch. Doppler measurements can also be ex-tracted but are omitted for now, see section 3.6.

The most important thing to realise is that the pseudorange measurements, essen-tially the range from satellite to receiver, are extracted from the C/A code by using thePRN code to it’s potential. This will be discussed further in the following sections.

The small correlation window of the public PRN code ingeniously allows the re-ceiver to calculate the distance to the satellite, providing a pseudorange. This rangecan be used to obtain a positioning solution. The word public in the previous sen-tence is quite important, since the military P(Y) code is not available for use in civilianreceivers. Since the code is needed to extract pseudorange measurements, no pseudor-ange information can be ascertained directly from the L2 signal. Only the secret P(Y)code is transmitted on that frequency, and not the generally available C/A code.

However, using several indirect techniques [16], exploiting the fact that the P(Y)code is modulated in the same way onto both the L1 and L2 signals, pseudorangescan be extracted from the P(Y) code on both frequencies. This brings the number ofobservables to three, one pseudorange from the C/A code and two from the P(Y) code.Note that a modernisation of the GPS satellite constellation is underway, adding theC/A code to the L2 signal as well. This will provide greater accuracy and robustnessand faster signal acquisition than the current L1 C/A-code signal [19].

Phase information can be extracted from both frequencies as well. This will be-come advantageous in later stages of this chapter, where the augmented positioningwill be clarified (section 3.4). Details on these observables will be discussed there.

Lastly Doppler related measurements can be obtained, again from both frequen-cies. The use of these signals will be discussed in section 3.6. For now it is clear thatin total seven observables are extractable from each satellite, each epoch. An extendedoverview of the observables is given in table 3.2. Note that for now we are only con-cerned with the pseudoranges from the C/A code and the Carrier phase measurements.

In the following sections, more emphasis will be put on how the measurementsare obtained, and how a valid position estimation may be calculated from them. In


Table 3.2: Observables available from a GPS satellite. L2C signal is being rolled out,but not yet completed

L1 signal L2 signal

Pseudorange from C/A code x -/xPseudorange from P(Y)-signal x xCarrier phase x xDoppler measurements x x

section 3.2 we will start this by presenting a basic method using only the C/A codepseudorange measurements. In section 3.4 more precise methods will be discussedusing the carrier phase measurements.

3.2 Standard Positioning Service

In the previous section the fundamental properties of the GPS system have been pre-sented. These can now be utilised in a method for standard positioning.

The main point is that the characteristics of PRN codes may be used for positioningpurposes. The PRN code generated by a satellite is linked to the atomic satellite clock.The code replicated by a GPS receiver is, in the same way, linked to a clock, howeverin this case, the much less accurate receiver clock. As stated before, by shifting thereplicated code in time and multiplying it with the received signals from the antenna,a satellite can become accessible to the receiver when the two codes align. If thecodes align, the receiver can compare the received satellite time, to the receiver time.The difference is the time it took the signal to cross the distance between satellite andreceiver. By multiplying this transit time with the speed with which the radio signalpropagates, i.e. the speed of light, one obtains the apparent range to the satellite, thepseudorange ρ (3.1).

ρ(t) = c[tu(t)− ts(t− τ)] (3.1)

Here ts(t−τ) is the emission time imprinted in the code and tu(t) is the arrival timeas measured by the receiver clock. τ is the transit time.

This pseudorange is however by no means the actual distance between the satelliteand the receiver. It is the result of a measured time difference, and several errors andbiases that influence this time difference are present in the signal. By rewriting (3.1),and making the most prominent of these errors and biases explicit, one obtains (3.2).

ρ(t) = r+ c[δtu−δts]+ Iρ +Tρ + ερ (3.2)

This is the core and basic equation for the pseudorange ρ. r is the actual range tothe satellite, the value one needs to calculate a worthwhile position solution. All theother terms are errors and biases that need to be handled and accounted for as much as

3.2. STANDARD POSITIONING SERVICE 23

possible.

The receiver clock for example is not nearly precise enough for sustained time-keeping, they are low cost clocks that are not synchronised with the much more pre-cise satellite clock. A simple approximate time is used in a receiver and hence a biasis present in the measured transit time. This is δtu. Note that the satellite clock toocan have a small bias with respect to GPS time (GPST). This is the time composed bythe control segment of the GPS system, used as the benchmark time to assess all GPSsatellites. This satellite time bias is δts. To convert these time biases to a range, theyare multiplied by the speed of light in vacuum, c. Other errors present in (3.2) are theionospheric delay, Iρ, and tropospheric delay, Tρ. These are present due to the fact thatthe radio signals propagate at varying speeds in different atmospheric layers. This hasan influence on the travel time. The last term ερ is a generic term for all other non-explicit errors still present in the pseudorange, such as receiver noise and multipath.These will be discussed later.

Now we can begin to piece together the final parts of the Standard PositioningService (SPS). We are now familiar with the concept of pseudoranges, but still thesecannot be used for positioning. First, errors and biases from equation (3.2) need to beeliminated as much as possible. Leaving these in as unknowns would make the result-ing equations unsolvable, ignoring them would result in a useless positioning solution.

Looking at section 3.1, one can already be quite successful in handling these bi-ases and errors. δts is broadcast in the navigation messages, allowing it to be utilised in(3.2). Furthermore an ionospheric model is broadcast through the navigation message,allowing an approximate ionospheric delay to be calculated in the receiver. This re-duces Iρ significantly. Moreover, if the receiver is a dual frequency receiver, receivingboth L1 and L2, the ionospheric delay can almost completely be removed, due to thedifferent effect of the ionosphere on both frequencies. The tropospheric delay is muchmore stable, and can be approximated by a model preprogrammed in the receiver. Thisallows Tρ to be significantly reduced. Measures to reduce ερ can include measurementsmoothing to reduce effects of multipath and measurement noise, however some resid-ual errors will remain present.

Now the largest remaining error in the pseudorange equation is the receiver clockbias, δtu. This bias cannot be corrected for in stand alone mode. Noting that thisreceiver clock bias is identical for all received satellites and hence all pseudoranges,this bias can be treated as a global variable, just like the ultimately sought after receiverposition x. Realising this, we rewrite equation (3.2) to (3.3).

ρ(k)c = ‖x(k)− x‖+b+ ε

(k)ρ (3.3)

‖x(k)− x‖ is the range of satellite k in vector form, where x(k) is the position ofsatellite k and x is the position of the receiver. b is the receiver clock bias convertedto metres and lastly ε

(k)ρ is the residual errors of the satellite, which differs for every

satellite and can, for now, not be minimised any further. By tracking for example ninesatellites, and extracting data from one measurement, or epoch, one acquires nine of


these equations, where x and b are the same for each equation, and x(k) is known fromthe almanac for each satellite. This leaves a total of four unknown variables (excludingthe ε

(k)ρ ) that need to be solved: three position values and the receiver clock bias. From

linear algebra it is known that these variables can be solved if there are at least fourequations, and thus if at least four tracked satellites visible.

The problem now is the fact that there are still errors present in the pseudorangesdue, among others, to ε

(k)ρ , not allowing for an exact solution of the problem. A so-

lution must be found that fits all the measurements best. This is commonly done byminimising the sum of the squared residuals, the difference between the pseudorangeof satellite k and the forthcoming calculated position. To do this we must first lineariseequation (3.3) by expanding it in a Taylor series. We may then use an approximate so-lution for the receiver position and clock bias and let the solution converge iteratively.Let us make a first estimate of equation (3.3) using initial estimations x0 and b0. Theequation then becomes:

ρ(k)0 = ‖x(k)− x0‖+b0 (3.4)

Now let the true position and bias of (3.3), x and b, be represented as x = x0− δxand b = b0−δb respectively:

ρ(k)c = ‖x(k)− (x0−δx)‖+(b0−δb)+ ε

(k)ρ (3.5)

Here δx and δb are the unknown corrections that need to be applied to our initialestimate. Using (3.4) and (3.5) we can subsequently develop a linearised equation(using a Taylor expansion) where δx and δb need to be determined:

δρ(k)c = ρ

(k)c −ρ

(k)0

= ‖x(k)− x0−δx‖−‖x(k)− x0||+(b−b0)+ ε(k)ρ

≈ − (x(k)− x0)

‖x(k)− x0‖·δx+δb+ ε

(k)ρ

= −1(k) ·δx+δb+ ε(k)ρ (3.6)

Now, considering an over-determined system, i.e. a system with more than foursatellites tracked, equation (3.6) can be written in a matrix form for K satellites. Wecan then use the least-squares solution to find the corrections to our initial estimates.This is represented in (3.7). [

δxδb

]= (GT G)−1GT

δρ (3.7)

With δρ the pseudorange errors of K satellites which can also be written as,

δρ = G[

δxδb

]+ ερ (3.8)

with G,

3.2. STANDARD POSITIONING SERVICE 25

G =

(−1(1))T 1(−1(2))T 1

......

(−1(K))T 1

(3.9)

These equations give the corrections δx and δb to the original estimates x0 and b0.We may now produce new, improved estimates using:

x = x0 +δx

b = b0 +δb (3.10)

Equation 3.3 can then be linearised again using the new estimates for x0 and b0,and the system can be solved again. This iteration may be continued until the changein estimates is small enough. This is usually within a few iterations, even with verypoor initial estimates.

In practise, a slightly different, but related, method is employed: Weighted LeastSquares. Up until now, it was assumed that every pseudorange has an equal validityand provides equally precise data: the variances of all measurements are assumedequal. This is however not the case. Therefore a weight matrix W is added to equations3.7 and 3.8. the result then becomes:[

δxδb

]= (GTWG)−1GTWδρ (3.11)

and:

δρ = G[

δxδb

]+ ερ (3.12)

The composition of this new weight matrix is outside the scope of this thesis, aswell as subject of research. However an indication will be given by clarifying the GPSantenna behaviour somewhat. The gain or amplification pattern of these antennas isnot equal in all directions due to physical limitations and design. For example it isgenerally not wanted that a satellite signal is picked up by the antenna that is comingfrom below the horizon. This is because this signal is in all likelihood a reflection ofa satellite signal bouncing of the ground. In other words a multipath signal, whichwill be described section 3.3. This signal can have a deteriorating effect on positionestimation and it is therefore not desirable to pick up these signals. An effect of theantenna design is that the resulting gain pattern is dependent on the angle the satellitesignal has relative to the GPS antenna. In appendix C the gain pattern an AeroAntennaAT2775-42 is given, this is the antenna that will be used in the Superbus system lateron. One can see there, that the amplification of a satellite signal originating form nearthe horizon, is much less than that of one originating from the zenith. This influencesthe signal to noise ratio of satellites that have a low elevation angle. This in turn causesthe pseudoranges extracted from these signals to have a larger standard deviation. This,as a last step, means that generally one wishes to give these measurements a lower


importance in the position estimation process. Arriving to this conclusion illustratesthe importance of the implementation of a weight matrix in the position estimationcalculations.

3.3 Measurement error sources

Now that is explained how a SPS solution may be found, and a possibility for moreprecise positioning is suggested, it can be useful to comment on the error sources,the reasons why the SPS solution will not reach the theoretical range measurementprecision. Some sources have already been briefly commented on, such as ionosphericand tropospheric delays, and methods for decreasing the resulting errors touched upon.Here some sources, now collected in ερ, that may have consequences for Superbus willbe considered. These are either modelling errors or stochastic errors. Modelling errorscan be improved upon by introducing better models. This is however not the case forstochastic errors. The main error sources mentioned here are:

• Control segment errors

• Receiver noise

• Multipath

Control segment errors are errors in the broadcast information, produced by thecontrol segment. These include the satellite ephemeris data, satellite clock bias andIonospheric model. These parameters need to be modelled and estimated by the con-trol segment apriori, and hence inherently have errors. Even so the prediction modelsare always improved, which in turn decreases the control segment error.

The signals tracked by a satellite receiver are always corrupted to a certain extentby the receiver. The noise being added to the received signals can come from a vari-ety of sources including, cables and antennas, but also thermal noise from inside thereceiver is a factor. Receivers are designed to closely follow the received signal andidentify the edge of a chip. Unfortunately, due to the fact that the edge of a chip isobscured by random noise, it is difficult for the receiver to identify the edge of eachchip. It is not a clear mark, forcing the receiver to make a decision on when the signalchanges. This introduces an error. In general, the receiver noise is dependent on signalstrength, which in turn is dependent on the satellite elevation angle.

Multipath is probably the greatest source of errors in a Superbus GPS system. Ra-dio signals show the same behaviour as light, in the sense that they can reflect whenencountered with an object. This is much like light hitting a mirror and changing di-rection. In figure 3.2 this is demonstrated. Here one can see that the radio signalsreflect of buildings causing the antenna to receive both the original and the reflectedsignal, but also merely the reflections alone can be received.

3.3. MEASUREMENT ERROR SOURCES 27

Figure 3.2: Graphical representation of a multipath situation. [source: BKG]


Signal reflections can play a major role in the Superbus due to the fact that it willroutinely drive in environments prone to signal reflections. This includes urban envi-ronments, but also for example sound barrier walls, commonplace near highways.

Typical errors in the pseudoranges differ for code or carrier phase measurements,with carrier phase measurement errors being about a factor 100 smaller. They canrange from several metres for code measurements to several centimetres for carrierphase measurements. Multipath can hence cause a serious effect in the positioningsolution of a receiver. This may be of great importance for Superbus when a precisionof a few centimetres is required, and due attention must be given to this fact.

The parameters mentioned above are also called nuisance parameters. These areparameters in the pseudorange equations that are not of direct interest (e.g. not positionand time), but do need to be estimated or eliminated. As was mentioned, estimationis never error-free and the elimination of these nuisance parameters would by far bepreferred. A table of typical measurement error sources and their effect on the pseu-doranges is presented in table 3.3.

Table 3.3: Typical error sources and their residual error sizes. [source: [31]]

Error source receiver type Potential error sizecode measurements phase measurements

Satellite clock 2 (m) 2 (m)Satellite ephemeris 2 (m) 2 (m)

Ionospheresingle frequency 1-5 (m) 1-5 (m)dual frequency 1 (m) 1 (m)

Troposphere 0.1-1 (m) 0.1-1 (m)Multipath 0.5-1 (m) 0.5-1 (cm)Receiver noise 0.25-0.5 (m) 1-2 (mm)

In the table the values concerning the carrier phase measurements are given aswell. Although carrier phase measurements are not explained in detail as of yet. Theirexistence is known however, and the nuisance factors associated with the carrier phaseare included as well for clarity and completeness. They will become valuable in thenext section.

3.4 Augmented positioning

In the previous sections the standard method of obtaining a PVT solution have beenrecounted. In this section the concept of augmented positioning will be described.It will become clear that more precise solutions are attainable in this manner. It willfurthermore become clear that this precision is not as easy to obtain as an SPS solution.

To begin this section, the carrier phase measurements must first be explored inmore detail. The precision gains will then become more visible. In the subsequent

3.4. AUGMENTED POSITIONING 29

sections, methods for using these measurements will be explained.

Carrier phase measurements

Most high end GPS receivers currently available have the ability to track the carriersignal, in addition to the conventional code tracking. The carrier signal is the physicalradio signal broadcast by the satellite as explained in section 3.1. This signal can betracked by a GPS receiver in a manner analogous to the PRN code. The receiver canreplicate sinusoidal wave of the tracked L band (1,2 or both). It can then shift thesignal in time, essentially shifting the phase of the replicated signal, until the receivedsignal and replicated signal are identical.

One might currently enquire what the benefits are of tracking the phase of thecarrier signal. This has to do with the tracking capabilities of the receiver. As wasmentioned in section 3.1, a GPS receiver is able to track a PRN code to about 1% of achip. For the C/A code this amounts to about 3 metres, therefore, without consideringother errors, the theoretical limit of the receiver. Modern receivers are able to track thecarrier frequency to about 1% of a carrier wavelength as well. A wavelength is con-siderably shorter than a PRN chip, 19,03 centimetres versus 300 metres. This brings avery significant precision gain, if the carrier phase could be used for positioning. Thecarrier phase can be tracked to about 1-2 millimetres. However some difficulties areencountered on the way to carrier phase positioning.

First we revisit the code pseudorange equations of (3.1) and (3.2). At first sight, itseems that the equation can be used for carrier phases as well. This is largely true, bytracking a carrier frequency the same problems remain as for code tracking. There arestill ionospheric and tropospheric delays, there are still biases in receiver and satelliteclocks, and the remaining errors are also still present. But when trying to accommo-date wavelengths in the pseudorange equations, one runs into a problem. The wholepoint of the receiver and satellite clocks was to obtain the range to the satellite bycomparing broadcast and reception times. This information is lost by only tracking thecarrier phase. But fortunately this is no great obstacle since both the carrier signal andthe PRN code are coupled to the same atomic clocks in the satellite. The receiver cangenerate the same carrier signal coupled to its own clock. The pseudoranges can thenbe deduced from the phase difference between the phase when it was generated, andthe phase when it is received.

Yet one final thing needs clarification. When purely looking at the carrier signalone ends up with precisely knowing the phase difference of the two signals, but noidea of the range to the satellite. The wavelength is about 20 centimetres, but therange to satellite is approximately up to 26500 kilometres. This leaves the user withprecise but ambiguous measurements, one does not know the number of cycles N thatpreceded the measured one. This indeed need to be added as an extra parameter to thepseudoranges leaving equations (3.13) and (3.14).

φ(t) = φu(t)−φs(t− τ)+N (3.13)


φ = λ−1 [r+ Iφ +Tφ

]+

cλ(δtu−δts)+N + εφ (3.14)

Here λ is the wavelength of the tracked L band. φu and φs are respectively thephases of the receiver generated and satellite received carrier signal. N is the numberof cycles preceding the measured phase cycle. The remaining parameters are analogto (3.1) and (3.2).

Introduction to augmented positioning

The building blocks for more precise PVT estimates with GPS are now becoming clear.With the help of carrier phase measurements a far more accurate estimation could bepossible. However the issue of the integer ambiguity remains. Furthermore, recallingsection 3.3, nuisance parameters such as ionospheric and tropospheric delays are stilla factor in phase measurements as well. They need to be reduced fairly drastically toobtain the cm precision needed for the Superbus. And even then, the solution needs tobe real-time to be of any value to the Superbus.

So if it is cardinal to remove nuisance factors, how can this be done? Simply put,there is one thing that all precise positioning techniques have in common. They willall use additional information to obtain solutions with a higher precision. Thereforethis group of systems is also called augmented GPS, they will use information fromoutside sources. They depend on this information for their precision, and thereforethey are no longer stand alone solutions. The extra requirement of a real time solu-tion only increases the difficulty of obtaining a solution, it does not change the factthat additional information is needed from a separate source. It only means that realtime communication between information source and GPS receiver should always bepresent in such a case.

In the next few sections, two options to obtain a precise PVT solution will bediscussed: Precise Point positioning (PPP) and relative positioning. The relative posi-tioning section will contain discussions on Real Time Kinematic (RTK and Flachen-korrekturparameter (FKP)).

Precise point positioning

Let us first look again to equation (3.14), repeated below. This brings back the core ofthe problem and serves as a suitable starting point.

φ = λ−1 [r+ Iφ +Tφ

]+

cλ(δtu−δts)+N + εφ [3.14]

Here λ is the wavelength of the tracked L band, r the range to the satellite, Iφ +Tφ

the ionospheric and tropospheric delays and δtu and δts represent the receiver and satel-lite clock errors. N is the number of cycles preceding the measured phase cycle, andfinally εφ are the remaining errors such as receiver noise and multipath.


Recall the way an SPS solution is obtained. Each tracked satellite provides onepseudorange equation, and four parameters remain identical within all these equations.These are the three position parameters (x,y,z) , embedded in r and the receiver clockbias (δtu). If we would try to solve these equations in the same way as we calculate anSPS solution, we would fail. There would always be more unknown parameters thanequations, since the ambiguity N is different for each tracked satellite: the system ofequations is under determined.

If we now recall the fact that once the ambiguity N has been found, it remainsconstant for as long as the satellite is being tracked. This in theory allows a PVT so-lution if those two conditions are met: the carrier phase ambiguities are known andthe satellite are tracked continuously. This is the approach precise point positioningtakes. However these carrier phase ambiguities cannot be solved directly. Again dueto the under determined nature of the system, the carrier phase ambiguities cannot besolved instantaneously. So rather than solving the system directly, parameter estima-tion algorithms are used to estimate the internal states (position, receiver clock error,and carrier phase ambiguities) until the system converges to a precise solution and thecarrier phase ambiguities N can be fixed.

Although this works in theory, it still does not offer an adequately precise PVT so-lution. As becomes clear from table 3.3, the remaining pseudorange errors are still fartoo large. Ionospheric and tropospheric delays still plague the pseudorange equations,as do the satellite clock and ephemeris errors. PPP therefore uses alternative ways tominimise nuisance parameters.

As was mentioned in section 3.2 the ionospheric delay can be largely negated byusing a dual frequency receiver. This is however not the case for any of the otherparameters. Therefore PPP uses alternative sources such as the International GNSSService (IGS) for superior estimation of the remaining errors. IGS is an alternativeand elaboration to the ground segment of the GPS system. It is a network of trackingstations with data- and analysis stations that monitor the GPS constellation and providesuperior satellite clock and ephemeris data, as well as superior broadcast troposphericdelay estimates in comparison to the estimates the SPS broadcast message provides.With the aid of the IGS data, PPP can generally provide centimetre level precision ona static receiver, while decimetre level is possible on kinematic receivers [3], both inreal time.

Limitations

PPP does retain certain drawbacks to SPS, making it, overall, unfit for use in the Su-perbus. First and foremost is the convergence period. The parameter estimation al-gorithms need a relatively long time to converge to a precise solution for the phaseambiguities. Furthermore the convergence time is dependent on many factors includedvehicle dynamics, where convergence is typically 30 to 40 minutes [6] -though signif-icant progress is being made [51]-, making PPP less flexible for use in the Superbus.Note that due to the presence of initial phase offsets, the phase ambiguities are not inte-


gers, but need to be estimated and subsequently kept constant as floating point numbersinstead. This significantly increases the difficulty of obtaining the proper carrier phaseambiguities. Though static convergence times are significantly shorter, this is not areal scenario for Superbus, which in essence is a kinematic GPS situation. Connectedto the convergence issue are losses of lock. The phase ambiguities are only fixed ifthe lock on the satellite remains fixed. Again this poses problems in kinematic re-ceivers. Superbus will encounter viaducts, aqueducts and urban environment, causinga GPS signal to be interrupted regularly, causing the need for a reinitialisation of theparameter estimation filter, an hence losing precise positioning. The long convergencetime is simply unacceptable in this scenario. The precise positioning will be ”out” fortoo long and, changes are, will never come to fruition due to the regular interruptionsin the Dutch landscape. Furthermore the decimetre precision of PPP is not quite incompliance with the 5 centimetre precision requirement stated for the Superbus.

Relative positioning

Another method to eliminate some of the nuisance parameters is relative positioning.A significant amount of the nuisance parameters turn out to be correlated in time andspace. This means that the errors change only slowly over time and space. By havinga reference receiver, or base station, on a known location close to the user receiver,or rover, a relative position of the rover compared to the base station can be obtainedwith substantial precision gains. Roughly speaking horizontal precision values of 1centimetre are attainable, depending on the distance to the base station.

The method is also called Real Time Kinematic (RTK) and computes the roverposition relative to a known base station, by sending raw observations, or correctionmessages to the rover (depending on the type of RTK). By doing this one can exploitthe temporal and spatial correlation of the ionosphere and tropospheric delays andnegate several nuisance parameters, enhancing the precision of the final positioningsolution to the centimetre level. The base station location is precisely known and thekinematic rover position is referenced to this location. If the base station is close tothe rover, for the moment let us assume a maximum of 50 kilometres, approximatelythe same satellites will be visible for both receivers. Note that when both receiversare tracking the same satellite, the satellite clock error (but to a large extent also theephemeris errors) of a satellite will be the same for both receivers. Add to this thefact that the rover position is related to the known position of the base station, and itbecomes clear that the error introduced by the satellite clock and ephemeris data willcancel. Also, the ionosphere and troposphere are highly correlated between the tworeceivers. The received signal at both receivers travels through roughly the same atmo-spheric conditions introducing largely the same atmospheric delays in the signal. Thisagain provides the possibility of essentially cancelling these nuisance factors. In totalthis provides the possibility of a very precise PVT solution.

As mentioned real time kinematic positioning can work by relaying time-taggedraw measurements of the reference receiver to the rover receiver, to obtain a PVTsolution. Let us look at the carrier phase measurements of one satellite visible by bothreceivers for a certain epoch.


Figure 3.3: Graphical representation of an RTK set up.

φu = λ−1 [ru + Iu +Tu]+ f (δtu−δts)+N + εφ,u

φr = λ−1 [rr + Ir +Tr]+ f (δtr−δts)+N + εφ,r (3.15)

Here the subscripts u and r denote the user and reference receiver respectively.Now let the pseudorange from the reference be sent to the rover receiver u. By differ-encing these pseudoranges the temporal and spatial correlated parameters drop out, aswell as the satellite clock error, see equation 3.16.

φur = φu−φr

φur = λ−1 [(ru− rr)+(Iu− Ir)+(Tu−Tr)]+ f ((δtu−δtr)− (δts−δts))

+(Nu−Nr)+(εφ,u− εφ,r)

φur ≈ λ−1rur + f ((δtu−δtr)+Nur + εφ,ur (3.16)

Assuming:

Tu ≈ Tr

Iu ≈ Ir

It is now clear that there are substantially less nuisance factors present in the equa-tion, only receiver noise, multipath and both receiver clock errors remain. This withthe condition that the measurements received by the rover are not too old and that thereference station is not too far away. These ”too old” and ”too far away” prefixes arefuzzy, since the correlation is dependent on atmospheric effects and time. As a rule


of thumb a degradation in horizontal precision of 1 centimetre every 10 kilometresdistance from the base station is taken. Time constraints on the received differentialcorrections are harder to determine, but they could be extrapolated to match the currentepoch for several minutes, again with degrading precision [25]. This degradation dueto time is the result of the fact that the RTK method has to transmit real time measure-ments from the reference to the rover receiver. Since this transmission inevitably takestime, the measurements have to be extrapolated, which invariably leads to degradationin precision.

The result is that we now have quite a precise carrier phase measurement, allowingfor positioning at the centimetre level. Yet now we once again have the problem thatthe whole system is under determined. Looking at equation (3.16) we can investigatethe number of unknown in the system. First of all we have the position vector, bringing3 unknowns (x,y,z). But now instead of one clock that needs estimation in SPS, wehave two, one for the base and one for the rover receiver. Luckily, these can be com-bined into one unknown. Lastly we have an integer ambiguity. Note that the positionvector and clock errors remain the same for each tracked satellite, yet the integer ambi-guity is different for each satellite. So expanding this for a system with K satellites, weobserve that there are 4+K unknowns in the system. Supposing we have a way to findout the K ambiguities they will stay constant as long as the satellites remain tracked.This would leave 4 unknowns in the system, which would bring the minimal numberof satellites needed for a solution, provided that the phase ambiguities are fixed, to 4as well.

Note that the Septentrio receiver used in the Superbus upholds a minimum of 5satellites for an RTK solution. 4 Satellites would mean that there is no redundancy inthe system. It is opted by Septentrio that this is not desirable, and only enables RTKwith a minimum of 5 tracked satellites. Secondly, Septentrio uses double differenc-ing, where base and rover clock errors together would cancel out completely, howeverthis is at the expensive of one visible satellite, again bringing the minimum number ofavailable satellites back up to 4 once ambiguities have been set.

This once again brings us back to the problem that the total system to compute aPVT solution is under determined. The ambiguities are still there, just like in section3.4. The problem does however get somewhat simpler. This is because the unwantednoise in the carrier phase measurements is lessened considerably. This is significantnot only for the accuracy of the final solution, but this also means that the ambiguitiesare now actually integers.

This brings us to the next step, The actual resolution of the ambiguities. Up untilnow it has become clear that the system of equations is under determined, but becomesdetermined, and hence directly solvable (epoch per epoch, with 5 or more satellites arevisible), once ambiguities have been fixed. For now we have an under determinedsystem of equations. In order to get to this next step, we therefore first have to re-solve these ambiguities. This can be done using the code measurements, or searchalgorithms (or both). These techniques however need multiple epochs to solve theambiguities [31]. This means that this method for precise positioning too (w.r.t. PPP)


has its drawbacks for use in the Superbus. Yet ambiguity resolution is attainable muchfaster (practically, within about 30 seconds), with a greater final accuracy after initiali-sation. Additionally these initialisation times are also possible in kinematic conditions.This fast calculation of the phase ambiguities, together with the centimetre accuracyobtained once they are fixed, considerably enhances the usability of GPS for high pre-cision kinematic navigation.

To conclude, the drawbacks of PPP are largely negated by using relative posi-tioning. The initialisation time is reduced to a great extent, to far under a minute inpractise. Furthermore, no convergence is present. Once the integers are obtained,centimetre level precision is possible. The fast initialisation also allows for more flex-ibility in the urban landscape, where some satellite lock may be lost momentarily.

Limitations

Though RTK is a promising candidate for the Superbus, able to deliver centimetreaccuracy at fast intervals (10 Hz), some problems remain. First of all note that noalgorithm for ambiguity resolution is 100% full proof, and an incorrect ambiguity can,in some cases, be fixed. In this case a bias will appear in the solution. Modern algo-rithms, such as the LAMBDA method [48], however do attain high success rates of99.9% when using L1 and L2, for so-called short baselines.

Secondly the ambiguity resolution is (practically) not instantaneous and problemscan arise from this. The Dutch landscape, open as it seems, has a considerable amountof constructions, blocking a clear view of the sky. Especially in urban areas, but alsoon highways in the form of viaducts, underpasses and noise barriers. These can mo-mentarily block all or enough satellites to lose an RTK solution. In other words, am-biguities are lost and need to be recalculated. Depending on the amount of satellitesthat can still be tracked, this takes from a few seconds to a minute, during which onlya SPS solution is present (which usually reestablishes within a couple of seconds af-ter reacquisition). This SPS solution is by far insufficient in terms of precision, andhence leaves the Superbus vehicle somewhat ”blind” during RTK reinitialising. Thisis a problem, especially in areas where blockages occur faster than reinitialisation iscompleted, leaving the Superbus lacking proper positioning for more extended peri-ods. This problem can be alleviated to a certain extent. First of all accurate velocitymeasurements are usually independent of positioning mode (RTK or stand alone), seesection 3.6. Heading information can be independent of positioning mode as well (seechapter 5). Furthermore, additional methods of positioning could be employed, for ex-ample by the incorporation of a Inertial Navigation System (INS), allowing the vehicleto be precisely tracked during GPS outages.

Another problem arising from the use of RTK for the Superbus is the roamingarea of the vehicle. As rover GPS receivers travel further away from a base stationthe relative position accuracy diminishes due to spatial decorrelation. Atmosphericdelays as well as constellation geometry start to deviate from the conditions present atthe reference station. For current generation carrier phase GPS receivers this is about


1 part per million (PPM) in horizontal direction, double that for the vertical. Thismeans that the standard deviation of horizontal position (σ) degrades 1 centimetre,every kilometre of distance (l) from the base station.

σtotal = σsystem +l

106 (3.17)

In addition the correction data from the receiver must not be too old. If RTK is tobe used on the Superbus, two new requirements need to be addressed:

• The need for constant data transmission to the vehicle

• The need for suitable corrections, able to accommodate the accuracy require-ments

These two items will be discussed in sections 3.4 and 3.7.

The mobility problem, FKP and VRS

As was mentioned in section 3.4, the precision of the calculated PVT solution de-grades with increasing distance from a base station. With the positioning requirementfor Superbus set at 5 centimetres, let us assume that a base station must be presentwithin 20 kilometres at any one time in order to make this possible. This 20 kmlimit is not a strict limit, and depends on atmospheric conditions, but is nonethelessrecommended by receiver manufacturers [42]. With the Netherlands being the initialplatform for the Superbus, this means that a significant number of base stations need tobe present, in order to get total national coverage. This is cumbersome and expensivefrom a infrastructural point of view. Stations and data centres need to be developedand maintained. The cost of such a network and its upkeep would be large, though,if necessary, theoretically well possible. In addition there are problems from the op-erational standpoint. Whenever the switch is made from one base station to another,the RTK algorithm needs to be reinitialised. This causes a gap in precise positioningof about half a minute, as was discussed in the previous section. If the Superbus is tocruise at 250 Km/h this means switching stations every 12 minutes, in a best case sce-nario. Theoretically a soft handover scenario could be possible, where the receiver willinitialise using corrections from the second station, before discarding the first station.This would however most likely require custom firmware from the receiver manufac-turer.

Luckily there are options to reduce the size of the base station network, while in-creasing the flexibility. This reduces the costs of the Superbus infrastructure whilemaintaining the 5 centimetre accuracy requirement. Two such options are using Fla-chenkorrekturparameter (FKP) or Virtual Reference Stations (VRS). FKP and VRSare two approaches to the same problem: minimising the distance dependent nuisanceerrors over larger distances than conventionally possible. This is the largest problemfor relative positioning.


The systems work by modelling the distance dependent parameters and relayingthese to the rover receiver. Consider an already existent network of reference stations.The exact positions of these stations are known, and they continuously receive satellitemeasurements. If enough of these stations are present, and enough measurements areobtained, one can model the complete state space representation of the complete GNSSsystem, and estimate all the parameters by using continuously updated Kalman filters.This parameter estimation can be very good, and can allow for millimetre accuratecarrier phase positioning if enough states are implemented [56]. The states of such asystem can include:

• Ionospheric delays

• Tropospheric delays

• Carrier phase ambiguities

• Satellite clock errors

• Satellite orbit errors

• Receiver clock errors

This basically means that every aspect of the pseudorange equations is containedand estimated, for each individual satellite, in the state space model. A completerepresentation of the GNSS system is present and estimated in real time, with greataccuracy. These factors can then be interpolated for every point contained within thisnetwork. Two ways to let the rover receiver benefit from this model will be discussedbelow.

FKP

The first way to better exploit the spatial correlation in a reduced base station networkis called Flachenkorrekturparameter [17, 56–58]. Figure 3.4 is a graphical representa-tion of the situation. Here two reference stations and a rover are represented, each at acertain distance from each other. The vertical axis is a representation of the error sizebetween the three receivers. One can see that in normal use of relative positioning abase station just sends out raw measurements of one point, ε1 or ε2. These correctionswill become less adequate with increasing distance from the actual reference site, ascan be seen in the difference between ε1, ε2 and the rover. FKP on the other hand pro-duces more information. It does send raw measurements of one single base station, butin addition it calculates a linear approximation of the spatial correlation of the errorsbetween multiple reference stations, the purple line in figure 3.4. Basically this allowsthe spatial variation approximation to be significantly better at a certain distance andhence produce a better solution than possible with the single base station RTK solu-tion. The RTK algorithm at the rover receiver uses the additional information and it’sown approximate position to calculate better error corrections.

The main advantage of FKP is that in principle no two way communication is nec-essary. This means that it can be broadcast, and rovers can operate while only receiving


data. However, looking ahead in the Superbus design, two-way communication willbe present in the Superbus. FKP can use this feedback of rough position estimationto optimise the FKP corrections sent to the rover. However, originally to conserve therequired bit throughput, this technique only sends out a simplified correction model tothe rover. Since ample bit throughput and two-way communication will be available,the following technique may be better suited for Superbus applications.

Figure 3.4: Graphical representation of relative positioning errors. [Source: 06-GPS]

VRS and PRS

This method can make more extensive use of the state space model, and is called theVirtual Reference Station (VRS) or Pseudo Reference Station (PRS) [12, 17, 54, 56].Virtual Reference Stations do not provide extra correction data in addition to the ob-servation data of the reference receiver. However, the state space model of the basestation network is used to calculate the prevailing nuisance errors near the rover andconsequently produce simulated observation data. This is then sent to the rover as ifit were a normal reference station. For this system to work two-way communicationis mandatory since the network must know an approximate location in order calculatethe errors and feign observation data. For various reasons, discussed in chapter 2 andsection 4.5, this will be the case for Superbus.

With this method, correction data for any location within the network can be cal-culated and sent out. This effectively means that distance errors due to the kinematicnature of Superbus can be circumvented, the VRS can just move along with the vehi-cle, as long as the vehicle remains in range of the reference station network.

Unfortunately, due to current practical limitations of the RTK algorithms, this isnot the case, the virtual station still needs to be fixed on one location, making a ”hardswitch” when relocating the VRS. This forces a reinitialisation of the RTK algorithm.This situation brings back the distance dependent error when the vehicle moves awayfrom the VRS, but more importantly it brings back the gap in the RTK solution forabout 30 seconds, when the VRS is relocated.

3.5. THE LAMBDA METHOD AND FUTURE ENHANCEMENTS 39

There is however a way to circumvent this. Instead of moving the VRS alongwith the rover receiver, one can place a VRS on an arbitrary location. Now instead ofsending observations from this location to the rover, the station sends out observationscalculated from the approximate position solution from the rover. In other words: thestation is static but the measurements it sends out are optimised for the position of therover and not dependent on the actual location of the base station. Another word forsuch a station is a Pseudo Reference Station, or PRS. This means that kinematic anddistance related problems do no longer exist, The corrections are always updated to thelast known rover position. In addition the base station switch can occur far less often,if ever. It is only limited to the maximum baseline length the rover receiver softwareprescribes.

The flexibility generated for kinematic applications by using VRS or PRS, in com-bination with the precision of the state space approach for the base station network,makes this an ideal solution for Superbus. Larger areas such as the Netherlands, canbe covered with greater precision, with less reference stations, than a network of in-dividual base stations. Also the problems arising from the frequent switching of basestations can be avoided. A last benefit from this approach is that commercial basestation networks are already operational. This completely eliminates the need for anyadditional infrastructure on the Superbus side, instead relying on an established net-work, and hence reducing costs.

3.5 The LAMBDA method and future enhancements

The LAMBDA method

In the previous section, the details of the method to resolve the carrier phase ambigui-ties have been omitted. Many current high grade receivers use the LAMBDA methodto solve the integer ambiguities. Although the method will not be explained in fulldetail here, since enough information exists on the subject [48], it will be explored fur-ther to allow some additional details to come to light. For a simple double-differenceexample the parameter estimation problem would be as follows [48]:

y = Bb+Aa+ ε (3.18)

where y is the observation vector, B and b the matrix and vector relating to the realvalued unknown parameters (e.g. position coordinates etc.). A and a are the matrixand vector relating to the double difference integer ambiguities.

The key concept of the LAMBDA method is to decorrelate, as much as possible,the usually highly correlated ambiguities of the n tracked satellite carrier signals, inorder to make it easier to estimate the integers correctly [53] [31].

The method is comprised of 4 basic steps. In the first step one disregards the factthat the ambiguities are integers and one estimates float solutions for the problem, b


and a. This is done using a least squares approach.

The second step is the decorrelation and integer estimation step. Because the floatambiguity estimators, a, are highly correlated, a simple rounding of the float solutionwould often times be wrong. In a 2D example, one may imagine the constant costellipsoid as very elongated (the ellipsoid where all a produce the same error from thecalculated float solution a). This is illustrated in figure 3.5. Here one can see in (a)that simple rounding to the nearest integers would produce an incorrect result. HenceTeunissen [48] proposed a transformation of the search space to a more decorrelatedone, in order to minimise the search space for the integer least squares estimator. Forthis to work the transformation matrix Z has to satisfy the following conditions:

• Z must have integer entries

• Z must volume preserving

• Z must aim to reduce the product of all ambiguity variances

By satisfying these conditions it is guaranteed the transformed ambiguities areagain integers, and that the variance-covariance matrix of the transformed ambiguitiesis more diagonal than the original [47] one.

A 2D example is given in figure 3.5 (b). Here the transformed search space isvisible, where a simple rounding would already produce the correct integers. TheLAMBDA method however normally uses the integer least squares estimator. Thismethod has been shown to be superior to other options, such as simple rounding orbootstrapping [46].

Step 3 involves the decision whether or not to accept the integers obtained fromstep 2. Several methods to decide to accept the integers are currently in use. The pop-ular ratio test will be discussed in the next section.

If the integers are accepted step 4 is simply to take the obtained integer ambiguitiesfrom step 2 and use these to correct the float estimates δb of step 1.

From a theoretical standpoint, the estimation method described here is optimal. Ofcourse, in practise, optimality is dependent on the method of validation used for step3. This will be discussed below. From a computational standpoint, it is more inten-sive than simpler methods, although in practise this has not been much of an issue,nor will it be likely to become one. High kinematic applications will certainly benefitfrom faster ambiguity resolution, but advances in the GPS constellation, or in integervalidation tests, are perhaps a better way to minimise the resolution time while main-taining a high success rate.

For example, both the modernised GPS, as the newly developed Galileo constel-lations, will feature 3 broadcast frequencies. The added third frequency will finallyallow for reliable single epoch ambiguity resolution (for short baselines) [33]. Also,

3.5. THE LAMBDA METHOD AND FUTURE ENHANCEMENTS 41

Figure 3.5: 2D example constant cost ellipsoids before (a) and after (b) the decorre-lation step. Rounding in the first scenario would produce incorrect integers of (4,5),while rounding in the second scenario produces the correct (−2,10). [Source: [31]]

using multiple GNSS combined will have much potential in reducing ambiguity res-olution times. Working on these solutions will be more effective than using anotherambiguity resolution algorithm.

The ratio test

It is clear that the results obtained from step 4 are most accurate if the integers foundin step 2 are actually correct. Remember that no integer estimation method has a100% succes rate, as the measurements are inherently noisy. For this reason many


GPS receivers use an acceptance- or validation test to maximise the chance that thecalculated integers are correct. An often used test is the ratio test. The test is definedas follows [49]:

Accept a if:q(a′)q(a)

≥ c (3.19)

Here a and a′ are the integer vectors of the best and second best solutions of theinteger minimiser q respectively (see [49]), i.e the closest to the float vector a. c is atolerance value to be selected by the user.

This means that if the second best solution for q is larger by a certain amount,controlled by c, a will be accepted as the final integer ambiguity solution. If this is notthe case the float solution will continue to be used until a solution is found for whichequation (3.19) is satisfied.

The method for accepting the calculated integer described here, is usable for min-imising the risk of accepting an incorrect ambiguity fix. There are however someproblems with this way of working, mainly due to poor implementation of the method.

Many people wrongly assume that equation (3.19) establishes a direct measure forthe correctness of the ambiguity vector a. If the ratio test holds, the solution will becorrect. This is however not true as the test only controls the chance (by choosing c)a wrong integer solution is accepted, i.e. it controls the failure rate of the ambiguityresolution algorithm.

Many receivers utilise a fixed value for c. Hereby they neglect the fact that the ratiotest itself is dependent on the prevailing GNSS model. In other words, a fixed valuefor c does not mean a fixed chance of an incorrect ambiguity resolution. This meansthat, with a fixed c, the end user has little control over what failure rate he or she findsacceptable.

So rather than using this fixed c, [49] proposes a variable number, in order to obtaina true fixed failure rate. For now this is not used in receivers and it obscures the viewon the failure rate of the algorithm. Additionally the value for c is not agreed uponin literature. The result of this can be that the algorithm failure rate is unnecessarilylow, causing unnecessary rejections of integer vectors by the test. This in turn causesthe time to first fix to rise. The suggested variable value for c truly enables an enduser to select a acceptable risk of a incorrect ambiguity fix, by (effectively) trading itagainst the time to first fix. It turns out that current receivers, with a fixed c are oftenon the conservative side with regard to the failure rate. This causes correct integer re-sultions to be disregarded by the ratio test and the system to stay with a float solution.The variable c can result in a more accurate failure rate, and therefore accept a integerresolution sooner, and hence improve the time to fixed ambiguities. This may be ofvalue for future GPS applications, not the least of which the Superbus, where it willbe shown that losses of lock will occur in normal operating conditions. Here effectiveand fast ambiguity resolution algorithms are paramount.

3.6. PSEUDORANGE RATES 43

The future for RTK

As is mentioned in section 3.4, the baselines of RTK may not become too large. This inorder to minimise the distance related errors, as well as to enable ambiguity resolutionwith a sufficiently high success rate. This is the main background for the FKP or VRSdebate above. When spatial decorralation become a factor, the ambiguity resolutionsuccess rate and the position estimation precision will drop below useful values.

In the previous section 3.5, is was stated that future developments may improveand simplify matters for Superbus. The onset of multiple GNSS and the addition of athird carrier frequency, will allow for reliable instantaneous integer ambiguity resolu-tion. But again, this is only true for short baselines, much like it is now. The onset ofwide area RTK might be interesting in this regard.

[33] and [13] describe a method for RTK that greatly increases the maximum base-line distance of current RTK. This Wide-Area RTK (WARTK) uses phase observationsfrom a network of permanent base stations to model the ionospheric delays for a largearea. This model can then generate precise and real time ionospheric corrections fora rover receiver anywhere within its network. The benefit is that this modelling of theionosphere can occur using a base station network with large spacing (several hun-dreds to over a thousand kilometres). This can enable fast ambiguity resolution on acontinental level. It will increase the operational theatre of the Superbus significantlywhile reducing the cost of the RTK network. Simulated Galileo data proved that an in-stantaneous ambiguity resolution success rate of 99.9% is possible at baselines of 114kilometres, while at 257 kilometres the time to first fix was 5 seconds. This comparedto a 0% success rate and a 75 seconds time to first fix respectively, when ionosphericcorrections were not used. Rover position estimates also remained below Superbusrequirements with 95% of estimates withing 0.8 centimetres horizontally and 1.6 cen-timetres vertically (ionosphere weighted with moderate signal strenght).

One can conclude from this that although RTK can adhere to Superbus require-ments using present day technology, the future will truly enable high precision kine-matic positioning in a large geographical deployment field. Instantaneous ambiguityresolution will become possible, while the baseline requirements for RTK will be-come more relaxed. Added GNSS constellations will make the RTK more robust aswell. GPS is interoperable with Galileo. This will make for more robust position es-timation since the loss of low elevation satellites due to signal blockage will be farless important due to the overall increase of the number of satellites. This increasein range measurements will also increase integrity of the whole system since integritymonitoring systems such as RAIM (see section 5.1) benefit from these extra measure-ments. Indeed research is already available on the subject and shows promising results[14, 15, 32]. More information on RAIM is available in section 5.1.

3.6 Pseudorange rates

One last subject that is of interest when using GPS for kinematic purposes are the pseu-dorange rates, or ρ. These allow the vehicle speed be tracked at considerable accuracy,


even in absence of an ambiguity fixed RTK solution. This can be of importance to theSuperbus project, for example when an RTK solution is unattainable. Therefore thepseudorange rates will now be discussed.

Recall the small remark in section 3.1 concerning the shift of the receiver repli-cated satellite signal. It was said that due to the PRN encoding, the satellite signalwould only become measurable if it was multiplied with an aligning replicated sig-nal. Only changes in time where hitherto considered, but due to the relative motionbetween a GPS satellite and a user, the observed frequency also changes, the Dopplershift. This means that in order to obtain a fix on a satellite, the receiver needs to shiftthe replicated signal in both the time and the frequency domain.

This Doppler shift however can in turn be used by the receiver to estimate veloc-ities directly, and not by differentiating between two position estimates. Indeed thismethod is used regularly within receivers, with good results.

The pseudorange rate, or relative speed between user and satellite is usually ob-tained from carrier phase measurements, and can be obtained by differentiating equa-tion (3.2), resulting in:

ρ(k) = r(k)+(b− b(k))+ I(k)+ T (k)+ ε

(k)φ

= (v(k)− v) ·1(k)+ b+ ε(k)φ

(3.20)

One can see that the receiver and satellite clock biases, now become the rates ofchange of these clocks, the clock drifts b and b(k) This has a much smaller effect on thesolution and so b becomes a nuisance factor, rather than an unknown. Furthermore onecan see that ionosphere and troposphere errors equally become subject to the rate ofchange. The rate of change of the ionospheric and tropospheric errors is usually quitesmall, minimising T (k) and I(k) , justifying them to be incorporated in ε

(k)φ

. Further-

more r(k) becomes dependent of the satellite velocity vector v(k), sent in the satellitebroadcast message, and the user velocity vector v, which has to be estimated. Lastly1(k) is the user-to-satellite line of sight vector, available through a user position esti-mate, or approximate position.

The user velocity can now be calculated in exactly the same way as a SPS solutionas described in section 3.2. First the the equation is rewritten to:

(ρ(k)− v(k) ·1(k)) =−1(k) · v+ b+ ε(k)φ

(3.21)

If we now simplify (ρ(k)− v(k) ·1(k)) to ˙ρ(k) we can write (3.21) for all the trackedsatellites as a system, in much the same way as section 3.2, to:

˙ρ = G[

vb

]+ ε

φ(3.22)

3.7. COMMUNICATION 45

Here G is the same (3.9), and v is the velocity vector of the user. Now in the sameway, using a least-squares method, a solution for the problem can be found.

The beauty of the method is that it produces quite precise results, since ionosphereand troposphere do not play a major role any more. Tests revealed a standard deviationof 0.015 m/s in horizontal direction and about double that in vertical direction.

3.7 Communication

In the previous sections it has become clear that a GNSS can in theory be used forSuperbus navigation, while adhering to the specified requirements. From section 3.4however it has become clear that in order to obtain a precise PVT solution, one last ob-stacle remains in place: Additional information from outside sources is always needed.This in turn means that in order to make precise GPS positioning work in a Superbusenvironment, the vehicle should have a permanent communications link. This mightnot be as straight forward as it seems in this culture of pan-communication. The vehi-cles speed of 250 km/h is an important factor in this. In addition, the Superbus teamis planning a (quasi)national deployment, which dramatically changes the way to ap-proach the communication system. This however will be the subject of this section.

Let us review consequences of RTK positioning for communication to and fromthe vehicle. RTK relies on a network of base stations to provide corrections for a largegeographic area. These observations need to be transferred to the vehicle. Considerthe VRS solution proposed in section 3.4. This solution requires observation data fromonly one reference station. This data is sent in a format set up by the Radio Techni-cal Commission for Maritime services (RTCM), called RTCM 2.3 [37]. Relaying theobservation data of 12 satellites in RTCM 2.3 requires a data rate of 6845 bps, or 6.8kb/s [50]. This figure is independent of the number of base stations present, and hencequite a stable bandwidth estimate.

Although this bandwidth requirement seems quite low, one must keep in mind thatthe data stream needs to be stable and constant. Recall the remarks of section 3.4concerning the RTK algorithm. If observation data is too old, for example due to anoutage, the algorithm will have to reinitialise, causing a temporary loss of RTK po-sitioning. For the communication this has several consequences. Data relay must beconstant, or at least so fast that no RTK time out will occur. System delays must bekept within acceptable levels for the same reason. Also, the envisioned vehicle speedof 250 km/h is not insignificant, and can render certain relay solutions, possible forstatic rovers uses, useless.

Although not the main subject of this thesis, a pragmatic solution must be foundfor data relay. Therefore some possibilities for mobile data transfer will be consid-ered here. Main advantages and drawback will be covered, and reflected against thecommunications requirements given below:

• Bitrate equal or exceeding 10 kb/s


• Able to relay data at speeds to a minimum of 250 km/h

• Latencies not greater than 2 seconds

• Uninterrupted data relay greatly preferred

Note that only two systems are regarded here. The section consists of a pragmaticapproach to the communications problem. This means that currently unavailable com-mercial systems are of limited interest. Although the Superbus concept is in search oftop-end solution, this does not mean that it should be a theoretical solution outside thecapacities of the project. When technologies mature this can always be reconsideredfor the concept. For now however, only systems within reach of the Superbus projectare considered, with information gathered from various sources [5, 20–22, 39].

Networked VHF radio

The first system that was proposed for Superbus data relay was a specifically designedsystem by the Koning & Hartman Company. During the initial phase of the Superbusproject Koning & Hartman was involved in the voice communication of the Superbus.A system was designed where a basic VHF system was used, which could in theory beupgraded to include data-relay.

The system architecture involves an elaborate network of VHF stations placed ad-jacent to Superbus routes. A station would need to be placed every 20 kilometres, dueto radio range constraints. This is an expensive and not very practical from an infras-tructural point of view. On the other hand, the need of a station every 20 kilometres,would open the possibility of placing a GPS reference station at the same location.This would decrease the dependence from third parties, such as commercial GPS basestation networks.

However the collaboration with Koning & Hartman was ended prematurely, anddetailed plans never became a reality. That said, the system never showed greatpromise for data relay. First of all the VHF system was designed from an analogstandpoint. It was initially designed for voice, and therefore the radios were not capa-ble of multiplexing. Voice was not converted to a digital signal before being sent. Thismeans that at one point in time only voice or only data could be transferred, with voicealways gaining priority. For RTK positioning, this is not acceptable. The problemcould be overcome by adding a dedicated channel for data, but this would again raisecosts, on both the vehicle and base station sides.

Yet this is not the only problem. Consider the system with a radio and referencestation every 20 kilometres. This can never be a viable operational system. In sec-tion 3.4 it was clarified that the carrier phase ambiguity remains constant as long as asatellite remains in view. Similarly, the observations from the reference station needto remain as well. If observations from a new reference station are received, the phaseambiguities change as well. This means the RTK algorithm needs to recalculate them,which can take up to a minute. With the Superbus obtaining velocities of up to 250km/h, this could mean the algorithm needs to be reset every 5 minutes. One can see


that the lack of an RTK solution for approximately 20% of the time, due purely tosystem architecture, is not a preferred situation.

Solving the above situation by using PRS or FKP could work, but then data centresare needed to compute the state space. Still the last problem, the handover betweenradio stations, remains. The radio receivers in the vehicle would only be able to beconnected to one radio station at any one time. If a switch is needed, the radio firstneeds to cut the connection from the first source, search for the second source andhandshake, before data relay can be resumed. The time this process takes was some-what unknown due to the speeds of the vehicle, but estimated to about 2 or 3 seconds.Though this should be fast enough, the loss of communication could again cause theRTK algorithm to reinitialise, losing an RTK solution for some time. This was how-ever never tested.

Considering all the potential problems and difficulties of the Koning & Hartmanradio system, it is dismissed for use as data relay system, though in theory it doesadhere to the requirements. The drawbacks of the system are too large. Development,construction and maintenance costs of such a system would be far to great to be of anyeconomical interest.

GPRS and UMTS

The conclusion of the previous subsection essentially forces us to search for third partysolutions for mobile communications. The cost of a (quasi)national private systemwould be considerable and, due to the very limited bandwidth to be used, simply noteconomically viable. In addition it would be a tremendous undertaking, which wouldtake years to complete. This realistically leaves commercial systems the only optionfor a cost effective national data relay system. This considerably narrows down theoptions and basically leaves GSM or the newly rolled out UMTS as candidates. Satel-lite communication is disregarded due to poor transfer speeds and high latencies. Itremains to be determined whether these commercially available systems are suitable,and satisfy the criteria stated in the beginning of the section.

General Packet Radio Service (GPRS) is an elaboration on the Global System forMobile Communication (GSM). GPRS and GSM are designated a second generationmobile communication technology, while Universal Mobile Telecommunications Sys-tem (UMTS) is deemed a third generation technology, currently taken into service.

Both systems operate in the Ultra High Frequency (UHF) band, with GSM mainlyon the 900 and 1800 MHz bands, and UMTS operating on a slightly higher frequen-cies, ranging from 1900 to about 2200 MHz. Both systems are cellular, meaning thatstatic cell stations are placed within an area, this is demonstrated in figure 3.6. Whena user roams across the network, the mobile device connects to one of these cells de-pending on signal strength, so in reality cells overlap. There are different sizes of cells,ranging from large base stations installed on masts, to small cells, placed on rooftopsin urban areas or even inside buildings. The size of these cells depends on different


factors, including geography, transceiver power and user density.

Figure 3.6: Cell structure of cellular networks. [source: [5]]

At this point the GSM and UMTS technologies start to diverge. GSM uses Gaus-sian Minimum shift keying (GMSK) for the modulation of data onto the carrier signal.This is represented in figure 3.7. This technology shifts the frequency of the carriersignal to accommodate the digital data signal. The frequency range of a cell is di-vided into smaller individual frequencies ranges of 200 kHz, allocated to individualusers. As an indirect result of modulation and the overlapping nature of the cells, eachadjoining cell in a network must use a slightly different frequency, in order to avoidinterference. This brings to light an inherent property of the GSM network. When auser is roaming, and moves away from the cell station, the signal strength decreasesrapidly, this in turn causes a switch to another cell. But since this cell operates in adifferent frequency range, the handover cannot be gradual. In other words, the switchto another base station causes a momentary loss of communication at the moment thecommunication transfers from one cell to another.

Figure 3.7: Representation of frequency shift modulation. [source: wikipedia]

UMTS on the other hand uses Code Division Multiple Access (CDMA) for modu-lation. This technology does not allocate each user a dedicated frequency, but instead


uses a larger frequency domain to transfer the data. CDMA is a form of spread spec-trum. The data is sent simply direct, spread across the entire bandwidth using a Pseudorandom noise code. In this way multiple users can utilise the full bandwidth a certainfrequency spectrum allocation allows. Diligent readers will now become alert, as thisis the same modulation method as used in the GPS satellite signal, discussed in section3.1.

Allowing multiple users to use a single communication channel without interferingwith each other is called multiplexing. This can be done in several ways. For example,a certain frequency domain can be subdivided into smaller frequency bandwidth seg-ments, much in the way GSM uses the bandwidth of a cell, called Frequency-divisionmultiple access (FDMA). Another way is to divide the bandwidth into time slots, andallow multiple users to use the bandwidth by accessing the signal at a certain sequenceof time intervals, called Time division multiple access (TDMA). CDMA on the otherhand, uses PRN codes to allow multiple user on one communication channel. Let usrecall the GPS satellite segment. In the constellation over 30 satellites send signalsat the same frequency while never interfering with each other. This is because of thePRN codes. The satellites send information to earth modulated in a certain code. Thereceiver replicates this code and multiplies it with the signals it receives. The interest-ing aspect of these PRN codes is that they are auto-correlating. In other words if thesignal replicated in the receiver and exactly aligned with the signal being sent out, thedata present within the signal can be extracted. In any other situation, where the twosignal are not synchronised, the two signals will cancel each other out. Please refer tosection 3.1 for more information on the subject.

In the case of mobile communication it works as follows. When a user enters acell, he will be assigned a specific PRN sequence. Whenever data is sent to the user, itis done modulated with this specific sequence, the PRN code. The receiver replicatesthis PRN code and multiplies it with the received signal. Remember that now only thematching received and replicated signals will correlate. All other codes will cancel outand therefore only the information destined for this specific user remains. In this waythe available bandwidth of a cell is used more efficiently. Multiple PRN signals can besent over the same frequency band and still every individual user is only able to extractthe correct signal without interference from other data being sent. One can loosely usethe analogy where a person is still able to comprehend and distinguish different lan-guages while in a airport lounge. It becomes a matter of tuning into the right referenceframe.

One might now enquire why this UMTS is deemed third generation technology,while it seems only a arbitrary step in modulation schemes. To explain this, it is firstimportant to uncover three inherent properties of mobile communication.

• Users are not bound to one geographical location

• Users can be on the move, e.g. have a non-zero speed

• Users solicit bandwidth


All these properties give rise to interconnected problems in mobile communica-tion. The bit rate of any transmitted radio signal diminishes with increasing distancefrom the transmission station.

This means that a mobile carrier must always have a network of cells to allow theusers to be truly mobile. The carrier must find a balance between bit rate and cell den-sity. It is evident that an increasing cell network density increases costs, and a reduceddata rate decreases usability. Furthermore, when using multiple cells, the user shouldbe able to switch between cells. This should cause no interruptions or dropped calls.These properties are also dependent on cell size. Lastly the user can have a non-zerospeed. This can in its own right cause problems in communication due to Dopplereffects. This again drops the data rate significantly with increasing speeds.

Now the differences between GPRS and UMTS can be explained. Firstly dueto the spread spectrum CDMA has a much higher native data rate than GPRS. Themodulation technique used in GSM, GMSK, is optimised for speech and low powerconsumption on the receiver end, and is therefore not ideal for the transmission of data.Also the GSM network will always prefer speech before data, meaning reduced or nodata transmission on busy cells as well as increasing latencies. Enhanced Data Ratesfor GSM Evolution (EDGE) has partly solved the low native bit throughput problemby using an alternate form of frequency shift keying, but it still relies on the GSMnetwork. The static bandwidth of the different systems depend on several parameters,such as the amount of fault correction used, but a general overview is given in table3.4. Note that the speeds of GPRS and EDGE can be improved by making use ofmultiple slots, this on the other hand directly reduces the number of users a cell canmanage.

Table 3.4: Theoretical data rates of mobile communication systems. [source:wikipedia]

Communication System Data Rate (kb/s)

GPRS 9 - 21,4EDGE 8.8 - 59.2UMTS 144 - 2000

One can see that all these systems can obtain the required data rate set for Super-bus, 10 kb/s. UMTS however, does this with a large margin, giving it more robustnessat speed. As mentioned, Doppler shifts can cause problems when a receiver is on themove. The frequency picked up by a receiver will become higher or lower than thetransmitted signal, when the receiver moves to or away from the source respectively.

Although the frequency shifts caused by a moving receiver are reasonably small,due to the high speed the signal has, it can have a pronounced effect on communica-tion. Especially GSM suffers from this due to the frequency modulation it employs.This modulation is by nature dependent on shifts in frequencies, and therefore has

3.8. CONCLUDING REMARKS 51

trouble dealing with velocities exceeding approximately 160 km/h, due to Doppler ef-fects. The number of bits that is incorrectly received (0 instead of 1 and vice versa)after transmission rises due to these problems, causing the need for the bit sequence tobe resent. This in turn causes the total data rate to drop. Furthermore, the frequencychannels can start to interfere with each other. GSM networks therefore employ guardbands between channels to reduce the chance of interference, but this in turn reducesthe total data rate of the cell.

UMTS and its CDMA modulation offers important benefits in this regard. ThePRN Codes are not dependent on frequency shifts and therefore the Doppler shift be-come a less prominent problem and is more easily handled. UMTS hence supportshigher receiver velocities, and in theory should support a data rate of 144 kb/s at 500km/h.

A further benefit of CDMA, and the independence of individual frequency chan-nels is the possibility for soft handovers. When a cell switch is needed, a hard ’hand-over’ is performed. The receiver must jump from one frequency to another. Thiscan cause problems, such as dropped calls and a interruption of data flow. SinceCDMA cells all broadcast over the same frequency, a hard switch is no longer re-quired, smoothly transferring the data flow from one cell to another. This eliminatesthe data interruption (important for sustaining an RTK solution), and reduces the prob-ability of a dropped call (depending on the cell overlap).

The pseudo random nature of the CMDA modulation also provides a better resis-tance to multipath than GSM. Multipath is the reflection of radio signals off objects,causing the signal to be received twice or more. This was introduced in section 3.3.Since the bounced signal will be slightly delayed it will have a bad correlation withthe receiver produced code, and hence become far less dominant or even cancel com-pletely. As an additional benefit the PRN codes are more private. Clandestine tappingof the signal is far more difficult, since prior knowledge of the used PRN code isneeded in order to pick up the right signal.

As a conclusion of this section, it is now possible to say that a permanent dataconnection with the Superbus is, in theory, possible. UMTS supports the requiredbandwidth, and more importantly, supports it up to the required speed of 250 km/h.UMTS is able to provide an uninterrupted connection across the complete networkwith acceptable latencies. Although the system is presently still not completely rolledout, it is in such a mature stage, that it will be usable, and will only get better. GSMin theory supports most of the requirements for use in the Superbus, and can be usedin absence of an UMTS but problems are expected above certain speeds, and possiblywhen switching cells.

3.8 Concluding remarks

In this chapter the basics of GPS positioning have been explained. The basic sig-nals, measurements and calculation methods have been clarified for practical use. The


Standard Positioning Service is clear, in addition to the more accurate methods of aug-mented positioning. It has become clear that the Superbus requirements, discussedin the previous chapter, are theoretically attainable. It is now also transparent thataugmented positioning is mandatory for the desired precision. With augmented po-sitioning comes the additional requirement of a wireless data connection, in order totransfer additional or correction data to the vehicle. UMTS is for now seems the mostpractical solution to this derived requirement.

Future practical implementations

With kinematic high accuracy positioning becoming easier and more accessible withthe onset of mobile broadband and base station networks, it has become a subject ofresearch to develop practical applications for road use. The accuracy obtained usingRTK is sufficient to start development of guided driver aids, or even autopilot func-tionalities in road vehicles. RTK is coming out of the laboratory phase.

Certainly this is of interest for Superbus, as it undoubtedly has a need for such sys-tems, in order to make the vehicle safer, faster and more useful. Perhaps, people arenot yet ready for high speed driverless transportation, but systems as lane assist, auto-mated emergency stops and active vehicle control in difficult situations are desirable,if only to improve safety. Several research programs in this area have been carried outalready, and provide an insight in the possibilities for Superbus in the not so distantfuture.

One such program is the development of a driver-assist system to enable bus driversto use the narrow road shoulder as an additional lane for high traffic roads [43]. Thenarrowness of the shoulder in combination with the width of the bus, provide a chal-lenging situation for any bus driver. The use of a base station network, an accurategeo-spatial database, and several on board sensors, are shown to make this kind of useof the road a safe effective possibility. An accurate road model provides a comparisonfor the real time position estimates, while several interfaces (head up display, hapticand tactile feedback) provide a way for the driver to keep the bus on the shoulder easilyand accurately, even in bad weather situations.

Of course, the scenario of an RTK outage as mentioned in section 3.4 cannot beignored. Underpasses or other unforeseen circumstances might interrupt the receptionof satellite signals momentarily, causing the integer ambiguities to be lost. In [43],a simple system using a commercial velocity and a yaw-rate sensor, or two velocitysensors, proved accurate enough (20 centimetres tolerable error) to allow the systemto continue to work for 15 seconds in the event of an RTK outage. Although this setup might not suffice for Superbus use, it does show, that such a system to cope withRTK outages can be developed at a reasonable cost.

Another article focuses on a more rigorous form of driver aid, building a systemto autonomously drive up a curvy road on order to study the vehicle behaviour at thelimits of handling [4]. In the end it might provide a system that is able to react and

3.8. CONCLUDING REMARKS 53

keep the vehicle in control when the driver makes an error. Where this system deviatesfrom stability control systems currently available in vehicles, is the fact that it can keepthe vehicle on a predetermined path.

The system is inspired by skilled racecar drivers, who have two major advantagesover a typical driver. Firstly they have an ability to predict vehicle grip more accuratelyand secondly they are able to use all vehicle actuators (including counter intuitive useof throttle and braking) more accurately to keep the vehicle in control in difficult situ-ations.

The system controls the vehicle into corners by estimating speed and steering com-mands, based on grip friction estimates and the desired path. When the vehicle hasentered the corner, the system uses the actuators (steering, throttle, brake) to cope withmodelling errors (for example initial friction estimates), and disturbances (deviationsin grip) to keep the car in control an on the desired path.

Again this may be an important safety feature for the Superbus. With an accurateroad model (already mentioned above), it may become possible to use this system invarious ways. It may for example be able to step in and override the driver in situationswhere the driver is about to lose control of the vehicle.

These two examples are only a way to introduce the possibilities that wide areaRTK enables in future vehicle and infrastructure design. For Superbus these technolo-gies may be very important, both with respect to safety and as its role as a technologydemonstrator. Section 4.5 will elaborate more on GPS related possibilities for the Su-perbus. This section only shows some direct research that is being done in the field.

In the next chapter a GPS system will be proposed for the Superbus. This systemwill adhere to Superbus requirements as much as possible. In subsequent chapters theperformance of said system will be evaluated.

Chapter 4

Superbus Positioning System

With the theory from chapter 3, and the requirements from section 2.4, all informationneeded to construct a GPS based navigation system for the Superbus is clear and avail-able. This is the goal of this chapter.

First a small recapitulation of the previous chapter will be given. Subsequently theNTRIP protocol, needed for the augmented positioning, will be discussed. The nextsection will cover the Positioning System, and how it is set up for use. Subsequentlythe Position module is discussed, this program allows for easy access of the GPS re-ceiver data to other Superbus systems. Lastly applications of the Superbus PositioningSystem are discussed.

4.1 Introduction to the Superbus Positioning System

In this section an introduction to this GPS system will be given. It is a small reca-pitulation of the possibilities and impossibilities, in order to clarify why the specificsystem in section 4.3 is developed.

Starting with the requirements stated in section 2.4, it is immediately clear thatthey exclude standard stand alone positioning. It is simply not precise enough. Thatsaid, SPS still can be useful as a back up to the precise system, but also for velocitymeasurements, as they are obtained through the pseudorange rates and therefore arealready more precise than the differentiation of two or more individual positioning so-lutions. This would allow some Superbus systems to remain functional when losing aprecise position fix.

This however only leaves augmented positioning as a possibility for the Superbus.This is in itself not a problem, but it does put extra strain on the additional requirementof (quasi)national deployment. The communications channel, needed for all types ofreal time augmented positioning, must be robust enough to be able to maintain therequired bit rate at the appropriate speeds, but it must also be able to do so for largeareas. This has been discussed in section 3.7. The conclusion there was that currently

55

56 CHAPTER 4. SUPERBUS POSITIONING SYSTEM

UMTS is the best solution for this problem. The network, although not thoroughlycomplete, is large enough for practical use and the bit rate exceeds the requirements.This could allow other subsystems and applications to use the data connection simul-taneously. This opens up a larger range of subsystems and applications that can makeuse of GPS, for example a networked database or other location based services. Anelaboration on this is found in section 4.5. It is also important to note that UMTS pro-vides the possibility of seamless transmission throughout the complete network, andshould support speeds, up to and exceeding that of the Superbus.

This leaves the decision of what specific augmented positioning system will beused for the system. Again looking at the requirements and the theory of section 3.4,one possibility is greatly preferred: relative positioning. At the time of writing, PrecisePoint Positioning is not robust and fast enough for practical use in the vehicle. RTK,with the use of base stations allows the precision requirement to be met, however onlyif close enough to a base station. This would mean the Superbus would only have adeployment area of roughly 10 to 20 kilometres around the base station. This mightbe sufficient for basic testing purposes, but will become less and less acceptable, asthe project matures. The solution to this can be the use of virtual or pseudo referencestations (VRS and PRS), that use a network of reference stations to compute local basestation observation values, which in turn can be sent to the vehicle. This has beenclarified in section 3.4

The very last hurdle is the mode of transmission of this observation data. This canbe done through the internet, using the NTRIP protocol, which would allow Superbuseasy access to base station information. More information of the protocol can be foundin section 4.2.

4.2 NTRIP

Network Transport of RTCM via Internet Protocol, NTRIP, is a standardised and easymethod of exchanging RTCM data, correction data for augmented positioning, overthe internet, making the system far more flexible than systems using direct data con-nections such as mobile radio transceivers [10, 40, 55].

NTRIP is based on the Hypertext Transfer Protocol HTTP/1.1. This means it is astateless generic protocol, which makes it very simple to use. Each request for RTCMdata is independent from previous connections, and therefore all relevant informationis transmitted in the request. This makes it a simple and lean protocol, relatively easyto implement on systems with limited resources.

The generic set-up of an NTRIP system is available in figure 4.1. Here one can seethat the basic correction or observation data is managed in RTCM, again a standard-ised protocol, but now for correction data itself. This protocol is flexible and allows forvarious methods of relative positioning, including the one prospected for the Superbus,RTK. These RTCM messages are available directly from an appropriate GPS receiver

4.3. SUPERBUS POSITIONING SYSTEM 57

and sent to a managing computer. This computer runs the NTRIP Server software.

The NTRIP Server then forwards the RTCM corrections it receives from the GPSreceiver to the NTRIP Caster via the NTRIP HTTP protocol.

The NTRIP Caster is the main body of the system. It is the program that organises,manages and maintains all aspects of the system. It compiles a list containing allnecessary information on the available NTRIP Servers and manages what informationis sent to an NTRIP client and if a client is authorised to receive this data.

The NTRIP Client is the end user part of the system. It acquires access to NTRIPServers through the NTRIP Caster. It also handles the user’s security aspects betweenCaster and Client, as well as the configuration of the desired corrections. The NTRIPClient delivers the correction data directly to the rover (user) GPS receiver depicted atthe bottom of figure 4.1.

With the NTRIP protocol one acquires a system which is independent of directdata connections, as was standard until recently. The system is more flexible as itallows multiple users on one reference station, and it is easily accessible as long asan internet connection is present. This is an ideal solution for Superbus granted apermanent internet connection is available.

4.3 Superbus Positioning System

The building blocks for precision GPS are now completely reviewed and explained.This section will contain the layout of the GPS system for the Superbus. This systemwill in theory be able to adhere to the requirements set by the Superbus team in section2.4. The overview provided here is a proof of concept. This is the system set up thatwill be tested in order to provide quantitative measures to establish usability in theSuperbus. An evolved version of this GPS system, more fit for quasi national roamingwill be discussed concisely, however no tests were carried out to support claims of us-ability. In theory this system should adhere to the requirements of section 2.4 as well,since no major changes are implemented.

In figure 4.2 it becomes clear that the system consists of two parts, a base stationsection and the Superbus section. The base station in the figure is placed preferablyon a accurately known location, in order to obtain the most precise results. It is conse-quently setup to output the necessary correction data for RTK positioning.

The communications between these two parts are standardised by use of the NTRIPprotocol. This protocol allows exchange of RTCM correction data through the internetas explained in section 4.2. This in turn means that the first part of the system as visiblein figure 4.2 is interchangeable with any other source that transmits RTCM correctiondata with the NTRIP protocol. This will become important during later stages of theSuperbus development, where the use of a single reference station might become in-sufficient. In this case a single base station can be substituted by a national network ofreference stations. This network will allow the use of Virtual Reference Stations (VRS)as discussed in section 3.4. An overview of this system is provided in figure 4.3. It is


NTRIP Server NTRIP Server NTRIP Server

NTRIP Caster

NTRIP Client NTRIP Client

RTCM correction data

from GPS receiver


from GPS receiver


from GPS receiver

HTTP HTTP HTTP

HTTP HTTPHTTP

HTTP HTTP

GPS receiver GPS receiver

RTCM RTCM

Figure 4.1: Generic NTRIP set up.


worth to stipulate that this switch to a networked base station system does not consti-tute any major changes at the Superbus side of the system as can be seen in the figure,as long as the data is available to the Superbus via the internet and the NTRIP protocol.

The second part of the system, the Superbus itself, is equal for both lay outs andworks as follows. The vehicle is equipped with a computer that has access to the inter-net via an UMTS connection. This computer is running several programs, int. al. theNTRIP Client software. This software either requests the RTCM data directly from abase station, as in figure 4.2, or it can request data from an appropriate base station,or a VRS station in a network, by sending the approximate (stand alone) position ofthe rover receiver. This is the case in figure 4.3. The applicable RTCM data is sentto the vehicle and relayed to the GPS Receiver e.g. via RS232 where it is handledas standard correction data. The receiver is now able to acquire an RTK position-ing solution and output this solution via TCP/IP. The output can either be in NMEA,sent onto the Superbus network ring directly, or the output can be in the SeptentrioBinary Format (SBF), a proprietary binary format, which can encompass more andmore precise data. Although the receiver can output NMEA with the same high pre-cision, the SBF output is preferred because the (standardised) maximum number ofcharacters per NMEA sentence (82) may be exceeded when high precision is selected.This will corrupt the NMEA sentence, rendering the valid PVT solution unreadable.Another benefit of using the SBF messages, is the additional information one can ob-tain. NMEA does not have standardised messages for all information; roll and pitchvalues are not represented for example, and only very basic heading information isavailable. The receiver solves this by adding a proprietary NMEA sentence. This log-ically negates the advantage of NMEA, namely the standardised nature of the protocol.

If the receiver is instructed to output information in SBF it must first be parsed toextract the required information before this information is sent onto the main networkring. The network ring is the main life line for many subsystems, problems in the ringwill upset all communication between subsystems. This obviously has large conse-quences for the vehicle and especially safety. Therefore access to the ring should beregulated and stressed as little as possible. Although the data throughput capacity ofthe ring would be large enough for a raw dump of all the SBF data, the fact remainsthat SBF is a proprietary format and hence should be decoded and parsed before therequired information is presented. Rather than dumping the SBF data onto the networkand letting each subsystem to extract the necessary information, another approach istaken. The SBF data is sent to a managing computer which is running the applicationthat will parse information output by the receiver, repack it, and only then redistributeit to the main network ring in the Superbus. This limits the amount of information be-ing released onto the network, and once the information is on the main network ring,it is freely accessible to every device and subsystem connected to the network.

Receivers and settings

Up until now the type of receivers and especially their output have been reasonablyimplicit. All information concerning this has been cloaked in amorphous terms: ap-


Gps Receiver

PolaRx2e

GPS Receiver

PolaRx2eh

HTTP

NTRIP Server

software

TCP/IP or RS232

GPS Antenna

NTRIP Caster

software

UMTS Receiver

GPS Antenna GPS Antenna

HTTP

RS232

Superbus

Base Station

RT

CM

da

ta

NTRIP Client

software and the

position module C++

codeTCP/IP

Superbus subsystems

Superbus TCP/IP Network

Figure 4.2: Superbus GPS system lay out with dedicated base station.


propriate, required, beneficial. This section will briefly discuss the receivers, and pro-vide more insight into the settings and output of these devices. Note that these settingsare only an initial starting point, and may need adjustments during testing to obtain thebest results with the complete system.

The receivers selected to power the GPS system for the Superbus, are developedby Septentrio. This is a relatively new company based in Leuven, Belgium. It is quitean open attitude company with close ties to the TU Delft. This allows us to interactmore easily with the company which makes it markedly easier to obtain informationand solve problems in this relatively new use of RTK positioning. Such relationshipswith other manufacturers have proven more difficult by the group of MGP. Key tech-nical details remain hidden from end users, which, for MGP, can be awkward at best.Septentrio has a more open dialogue, whilst offering a state of the art approach to GPSreceiver development [44].

The PolaRx2e is selected as the base station receiver, for the initial stages of de-velopment. This is a 32 channel, single antenna, dual frequency receiver capable ofproviding correction data in both RTCM 2.3 and RTCM 3.0. The antennas also stemfrom Septentrio, the PolaNt, though they are rebranded dual frequency antennas fromAeroAntenna.

The output settings of the reference receiver are predominantly kept on default basestation settings. However to obtain the best results the position of the phase centre ofthe antenna must be input into the receiver. All testing to validate the GPS system inthe next chapter has been done with one base station, RTCM 2.3 and the AeroAntennaAT-2775 antenna.

For the vehicle, the PolaRx2eH is selected. This is largely the same receiver withthe exception that it serves two antennas, again from the type AT-2775. This allowsthe receiver to set up a short baseline between the main and the auxiliary antenna, andhence provide accurate heading and pitch information. The receiver can accept a mul-titude of correction data over various ports, and subsequently output an accurate RTKPVT solution, using the LAMBDA method for ambiguity resolution. Full specifica-tions of the receivers can be found in appendix A. The PolaRx2 series receiver and theantenna are represented in figure 4.4.

The settings of the receiver can be found in table 4.1. The elevation mask hasbeen set at 10 degrees due to the installation of the antennas in the Superbus. They areplaced in carbon fibre, e.g. highly reflective compartments, and care should be takento adjust this setting once installed. The antennas should have a view of the sky that isas large as possible, while the multipath resulting from the reflections of the compart-ment should be limited. Note however that the elevation mask should not be smallerthan the 10 degrees currently set because low elevation satellites are far more proneto multipath and losses of lock. This influences the robustness of the RTK solutiongreatly. The placement of the antennas in the vehicle is quite important as well. Sincethe auxiliary antenna position is calculated using a small baseline, the antennas shouldhave largely the same aerial view, so that the same satellites are tracked. Also the


GPS Receiver

PolaRx2eh

NTRIP Caster

software

UMTS Receiver

GPS Antenna GPS Antenna

HTTP

RS232

Superbus

NTRIP Client

software and the

position module C++

codeTCP/IP

Superbus subsystems

Superbus TCP/IP Network

Base Station Base Station Base Station

State space model of the

base station network.

RT

CM

da

ta

Ap

pro

xim

ate

po

sitio

n

Base station Network

Virtual Reference

Station

Figure 4.3: Superbus GPS system lay out with network based VRS.


Figure 4.4: Septentrio PolaNt antenna and the PolaRx2eH receiver. [source: Septen-trio]

Receiver Settings

PVT Output 10 (Hz)Measurement Interval 0.1 (s)PVT Mode RTK and Stand aloneRTCM Input RTCM 2.3Input Port Com 2Satellite Tracking GPS onlyTracking mode DynamicReceiver Dynamics HighElevation Mask 10

Measurement Fit offSmoothing Interval 100 (s), 10 (s) InitialisationRemaining Settings Default

Table 4.1: Provisional Set up of the PolaRx2eh receiver

longer they are placed apart, the more precise the heading precision will become. Inthe Superbus the antennas are placed centrally in the roof along the longitudinal axisof the vehicle, and are placed 8.19 metres apart. This was the farthest apart they couldbe placed, while maintaining a similar view of the sky. A situational sketch is providedin figure 4.5.

The tracking sensitivity is set on medium, as a trade off between precision androbustness of the solution. To accommodate the dynamic nature of the vehicle, thetracking mode is set to dynamic. On the PVT calculation side, the receiver dynamics


are set to high. This is done to help the Kalman filter to filter out measurement errorsadequately. Furthermore the PVT interval is set to 10 Hz.

Code smoothing is used to reduce pseudorange noise and multipath. Carrier phaseand Doppler measurements are unaffected by this. The smoothing interval is 100 sec-onds, with a 10 second alignment. During this alignment the smoothed code is notused for the PVT calculations. In the Septentrio receiver dual frequency carrier phasemeasurements are used to smooth the code measurement. This allows long smoothingintervals to be used without any negative side effects.

A measurement fit is not used. Normally the measurement interval of the receiveris greater than the output interval. The measurement fit would fit a quadratic poly-nomial over all the measurements within the output interval and evaluate these withthe measurement interval of the output epoch. This will reduce multipath, but at theexpense of a latency. This is not desirable in this highly dynamic application of GPS,furthermore the measurement interval is chosen equal to the output rate of the PVTsolution.

To conclude the list, Satellite Based Augmentation System (SBAS) satellites arenot tracked, due to the fact that two dual frequency antennas are fitted. The trackingof SBAS satellites would use up valuable receiver channels. Without tracking SBASSatellites, the receiver can track up to 8 satellites on two frequencies, on each antenna.This number should not be reduced, since the receiver will only start the initialisationof an RTK solution when tracking 6 satellites, on two frequencies. Additionally, track-ing these satellites would not benefit the Superbus. The accuracy of the SPS solutionsusing SBAS is not sufficient for use on the Superbus.

The set up that was used during testing, figure 4.2, serves as an initial startingpoint for further testing in the Superbus itself. Additional settings could be needed,and some settings might become obsolete. For example, one setting is for now omit-ted, but can become very useful once the receiver and antennas are permanently fixedinto the vehicle. The location of both antennas can namely be fed into the receiver.If the antenna geometry is known, the receiver can use this information to validatethe calculated integer ambiguities. This can accelerate time to first fix values, and thedecrease the probability of a wrong ambiguity fix, especially in single frequency op-eration [41]. The settings of table 4.1 however should give a solid starting point forfurther testing and tweaking in the vehicle itself.

4.4 The Position Module

In section 4.3, it has become clear that the GPS receiver in the Superbus vehicle is notdirectly connected to the main network, but instead is routed to a computer first. Thisallows a computer program to act as a buffer and controller between the network andthe GPS receiver. This program, the Position Module, has two main tasks:

• Set up communications and initialise the receiver

4.4. THE POSITION MODULE 65

Figure 4.5: Placement of the two GPS antennas in the Superbus.


• Process the output SBF data before sending it to the main network

One advantage of the system set up as shown in figures 4.2 and 4.3 has alreadybeen mentioned in section 4.3: it limits the amount of data on the main network ringby extracting all the desired information from the SBF messages, and repacking it inone simple message, before sending it out on the main network. This will simplify theway all the subsystems must access GPS data as well.

However several more benefits are present. Most importantly: the program con-trols the in- and output to and from the receiver, as well as the output to the mainnetwork. The Position Module stops all the data of the main network from reachingthe receiver. This is desirable since it eliminates the chance of an accidental input inthe receiver. The receiver listens to everything coming in over the TCP/IP port and willrespond accordingly if a correct input argument is presented. Although the chances aresmall, an accidental string of characters from another subsystem could cause an un-intended reaction from the GPS receiver. Since the computer that runs the PositionModule has two TCP/IP ports, one for the GPS receiver and one for the network ring,data from the network ring does not reach the receiver automatically, eliminating theaccidental input problem.

The Position Module also controls the settings of the receiver. Although they canbe saved internally, the program re-initialises the settings every time the program isrun. This makes it a more robust solution, in the initial stages of testing. In the eventof a problem with the network ring, GPS receiver or any other subsystem in the vehi-cle, a simple restart of the Position Module could allow the GPS receiver to connect tothe network again while outputting the correct data. Cutting power to the receiver andthen letting the system reboot can achieve the same result, however with a significanttime loss.

The final task of the Position Module is the processing of the SBF block, extractingthe appropriate data, and redistributing it through the main network. This is an infiniteprocess and will continue as long as the receiver is operational.

Program structure

In this subsection the general lay-out of the code will be presented. The PositionModule has two main tasks as already mentioned above. First it is to set up a com-munications channel with the receiver. When the channel is opened successfully, theprogram initialises the receiver by sending all the necessary settings to the receiver.This includes the general settings mentioned above, but also the SBF blocks to be out-put. When this has been successful, the Position Module switches to the other task athand, to interpret the SBF output, extract the relevant information, and package thisinto a single package. This package can then be sent onto the main network ring.

In figure 4.6 a schematic overview of the program is given. This will now be ex-plained in more detail. The program starts by declaring all dependencies, subroutines


Figure 4.6: Position Module, high level lay out.

and global variables. This is standard practise and needs little explanation. The Posi-tion Module then enters the main routine, where it will first perform the initialisationof the receiver. Local variables are declared and the initialisation subroutine is called.This subroutine will itself perform several tasks. It will first open log files that willsave all the information that is output by the receiver. This is mainly done for debug-ging and could be removed if deemed necessary. A serial connection with the receiveris subsequently established. The connection is set to its maximum speed to obtainenough throughput for the output: 115200 baud, 8 bytes, no parity and 1 stop bit. Ifthe connection with the receiver is successful, the proper settings, as discussed in table4.1 are sent to the receiver. In addition the command is issued to output several SBF


blocks. These blocks will in total contain all relevant parameters necessary to operatethe GPS system in the Superbus. A list of these parameters can be found in table 4.2.

When the initialisation has been completed, the program enters the process loop.Here all the incoming SBF blocks will be read and parsed for the relevant parameters.Each SBF block is composed in the same fashion by the receiver. It is a binary se-quence which will always be composed of a multiple of 4 bytes. The message itselfopens with a header followed by the body. Padding bytes are potentially used as fillingto complete the message. The header commences with a synchronisation field, ”$@”,used to identify the start of an SBF block. The header then contains a cyclic redun-dancy check (CRC) to detect any errors resulting from the data transfer. Then messageidentification is given, a unique number discerning each type of SBF block. Lastly themessage length is contained within the header.

The information contained in the header is used in the processing of the SBF blocksas follows. The incoming bytes are stored in a buffer which is searched for the syn-chronisation, ”$@”. When this is found the subsequent 6 bytes are read as well. If thesynchronisation characters were not random and the message was convoyed correctly,these 8 bytes must constitute the header of an SBF block. A first check for this iscarried out by checking the last two bytes of the header. If all is well this will be thelength of the SBF block. This must always be a multiple of 4 bytes, always positiveand always less than 4096. If this is not the case, the message is discarded, and thesearch continues for the ”$@” synchronisation. If the last 2 bytes do uphold the pre-vious requirements, the program moves to the next stage. The program runs numberexpressed in the 5th and 6th bytes of the alleged header through a list of SBF blocksthat contain relevant information for the Superbus. If the number does not pair up withone of the preprogrammed blocks, the program will again return to the search for thesync. If the number is contained in the list of relevant SBF blocks, another scenariounfolds.

The length and information contained in an SBF block is well documented, andavailable in the manual [41]. The program will compare the length of the SBF blockcontained in the header, with a hard coded length, the length the block should be. Ifthese numbers do not coincide, the search for sync is again initiated. If they do co-incide however, the Position Module will read out said number of bytes and fill theparameters contained in the SBF block, preprogrammed in the Position Module, withthe values provided by the received bytes.

An example: the first four bytes of the message body of an SBF block usuallycomprise the time of week. The data contained in these four received bytes are nowassigned to the time of week parameter of the SBF block in question. The next twobytes are usually the week number, these bytes are assigned to the week number of theSBF block. This reading and assigning of values continues for all parameters in theSBF block. If the block parameters are filled with values, a CRC check is performedand the data is output to the log files. Within the program itself, a flag is also set toshow that the SBF block has been filled. If enough blocks are filled with data fromthe same epoch, the module continues with the next step. If this is not the case it will


again move to the beginning of the process loop, the search for ”$@”.

The next step is simply to fill a new structure, GPS data. This structure containsonly parameters relevant for the Superbus, and this is filled with values obtained fromthe collected SBF blocks from the previous step. The parameters of GPS data are rep-resented in table 4.2. Once it is filled, the GPS data block can be sent onto the networkwhere the data is easily accessible for all Superbus subsystems. This part of the pro-gram is not functional yet, and needs to be completed before the Position Module canbe used in the vehicle. The end of the loop is now reached and the program will againstart the search for the synchronisation. The process repeats, the stored SBF blockswill be overwritten with new data until GPS data can again be occupied with new data.

Output

The last subject of this section will involve the output of the receiver. What are theparameters in the assembled structure, GPS data, and why are they selected? As men-tioned in the previous section, the selected parameters can be found in table 4.2.

Position Module Output Parameters

PositionLatitude

LongitudeAltitude

Velocity

GroundspeedVn

Ve

Vu

TimeTime of WeekCurrent Date

Current Time in UTC

AttitudeHeading

PitchCourse over Ground

Integrity and Safety

PVT Mode NrSVIntegrity Flag PVT Error Flag

PDOP TDOPHDOP VDOP

HERL Position VERL PositionHERL Velocity VERL Velocity

Distance to Base Station Age of Last Correction

Table 4.2: Provisional output for the PolaRx2eh receiver


The parameters output by the Position Module are basically divided into five parts:Position, Velocity, Time, Attitude and Integrity and safety. An example output canbe found in appendix D. Most of the parameters contained in these sections are quitestraight forward and need little justification, but all will be covered shortly.

Latitude, Longitude and Altitude (above ellipsoid) in the Position section are givenin WGS-84 while in stand alone mode, in RTK mode, the reference frame of the basestation coordinates is used. However, the WGS-84 is still used for calculation of thebaseline, since the satellite orbits are given in WGS-84. The velocity outputs, Vn, Ve,Vu, are those in the north, east and up direction. A ground speed is also provided, asimple resultant vector of the three directional velocities. The time is given in Time ofweek (TOW) and in UTC Time. The current date is also presented. Attitude aspectsof the GPS data structure are the heading, pitch and course over ground. The first andthird respectively may occasionally not coincide, with heading being the true attitudeof the two antennas, while the course over ground (COG) is the track heading of thevehicle.

The last part of the output consists of the Integrity and Safety information andmight need some more explanation. The PVT Mode parameter provides informationabout, logically, the PVT mode in situ. This is important information as it gives an easyand direct indication of the maximum precision that can be expected from the receiver.While in stand alone mode, certain subsystems in the Superbus may want to decidenot to use certain PVT information due for example to lacking precision. Subsequentlythere are two parameters that are flagged only if the system is not working as intended.These parameters are the PVT Error Flag and the Integrity Flag. The PVT Error flag israised when a PVT solution can not be attained. The flag can differentiate 9 differentscenarios, as can be seen in table 4.3.

Table 4.3: differentiations in the PVT error flag

PVT error scenarios

1 Not enough measurements2 Not enough ephemerides available3 DOP values too large ( 15)4 Sum of squared residuals too large5 No convergence6 Not enough measurements after outlier rejection7 Height greater then 18km and speed greater than 515 (m/s)8 Not enough differential corrections available9 Base station coordinates unavailable

This can be very useful for debugging, or again for subsystems to decide whetheror not to use certain information being output. The Integrity Flag is produced by theRAIM algorithm, discussed in section 5.1. The settings for the RAIM are configurable,so the user can influence the circumstances in which a flag will be raised. Needless tosay, integrity is compromised when the flag is raised, in which case PVT data should

4.5. GPS APPLICATIONS AND FUTURE DEVELOPMENTS 71

not be used or with extreme care. For certain applications, more stringent criteria thena simple flag might be appropriate, This can be programmed into the software at thesubsystem side, by using the External Reliability Levels (ERL’s). These parametersgive a quantitative measure of reliability of position or velocity solutions. A positionor velocity solution passes if the value is lower than a preprogrammed alarm level atthe subsystem side.

The Dilution Of Precision (DOP) values output, are another way to interpret theGPS solution. More precisely, it provides information about the effects of the currentsatellite constellation geometry. Although not very comprehensive, the DOP valuescan be useful to quickly overview the current GPS situation. Generally PVT solutionswith values over 6 for the Position DOP, PDOP are considered not usable. Extensiveelaboration on DOP’s is outside the scope of the thesis. However, a good introductoryarticle is available in the bibliography [18].

The last three parameters are the number of satellites in view (NrSV), the distanceto the base station and the age of the last correction. The NrSV is always a quickbut coarse way to assess the robustness an precision of your PVT solution. The moresatellites available the more robust your solution will be: the system will more likelybe able to cope with variations in operating environment without loss of functionalityor failure. The Distance to Base Station can be an important parameter as well. Asindicated in section 3.4, the precision of relative positioning decreases with increasingdistance to the base station. This parameter therefore can provide an extra indicationof the state of the RTK solution. Note however that while using VRS for correctiondata, this may not be the case. These stations might be placed at considerable distancefrom the vehicle. The corrections received are however calculated for distances muchcloser to the Superbus. The Age of Last Correction is provided for somewhat the samereasons as the Distance to Base Station. In the same section 3.4 it has become clearthat correction data may not be too old, or it will become useless. For the system inthe Superbus, the maximum age of the correction data is left at the default value of theSeptentrio receiver, 20 seconds. Under normal circumstances the latency will neverbecome this large and therefore this parameter can provide circumstantial informationabout the robustness of the radio connection and hence your RTK solution.

4.5 GPS applications and future developments

In this chapter the Superbus GPS system has been clarified. The system, the settings,and the output are now clear. It is now time to take a closer look at the purpose of thesystem.

The intended use of the GPS system is to provide information to other Superbus(sub)systems. These systems exceed straight forward uses such as simple navigationsoftware. The data can also used for more elaborate systems. In this section the pos-sibilities of the Superbus GPS system will be explored in more detail. This includeshardware systems and software systems, for both immediate use as for future develop-


ment.

Hardware

In the early design phases of the Superbus project a precise positioning system wasto be used by at least two hardware subsystems, the active suspension system and theradar system. These will be considered in more detail below. In the process a valida-tion for the GPS requirements will come to light.

Active suspension would allow a vehicle to stiffen and soften the individual wheelsuspension to alter the vehicle behaviour in different circumstances. An example canbe seen in figure 4.7, where the (driver-) right side suspension is stiffened to counteractbody roll in a turn. One can clearly see that the vehicle with active suspension (right)is much more level while cornering. This improves handling and passenger comfort.For Superbus this system of actively controlling all wheels can be of particular inter-est. Consider a scenario where the suspension would be significantly softened whenpassing a small identified bump in the road. The vehicle would then easily glide overthe imperfect piece of infrastructure due to the relatively low weight of the wheels,without the need to slow down to the extent normally required. This alleviates the re-quirement for a high cost, perfect and smooth highway, and so reduce the overall costof the Superbus infrastructure. The active suspension also increases passenger comfortin the same process.

The system can be expanded to different scenarios as well. For example while cor-nering and in city traffic. During corners the ride can be smoother by reducing bodyroll. City traffic can be made more comfortable by reacting to speed bumps and othercity obstacles. When the chassis is raised during city travel, the centre of gravity isalso raised. This aggravates the body roll in turns and as well as the reaction to speedbumps. An active suspension can react to this change of vehicle dynamics. This trainof thought can be continued to overall handling as well, improvement in this sectionimproves overall safety of the vehicle.

Figure 4.7: Vehicle behaviour during cornering. Active suspension is enabled on theright. [source: Bose]


For the active suspension to work properly, two main points are important. Thefirst would be a flexible, and highly up-to-date database of the Superbus infrastructurewherein all locations of road imperfections could be stored, adapted, and accessed.The second would be a reliable and accurate positioning system, that could facilitatethe possibility of a fast and accurate alteration of the suspension. Hence the decisionfor the 5(cm) accuracy, and 10 Hz output rate. The first Superbus has however beenbuilt without active suspension due to reliability and safety concerns. A system failurein the higher speed regions could potentially lead to a hazardous situation. For thisreason the integrity and availability of the GPS system PVT solution needs to be ashigh as possible. The active suspension was deemed not mature enough for immediateuse, but remains of interest. The requirements for the GPS system remain standing forthis reason, the system may still be evaluated for further development of the Superbus.

The second subsystem that was to benefit from an accurate positioning system isthe radar system of the Superbus. Radar image processing is heavily dependent on theangular velocities of the vehicle. High accuracy, high frequency angular velocities areneeded in order to properly extract data from the collected radar reflections. With adual antenna GPS set-up as is now built in the vehicle, these angular velocities couldtheoretically yield sufficiently accurate results. Unfortunately two factors discouragedSuperbus to implement this approach. Firstly, the PolaRx2eH receiver currently doesnot output angular velocities directly, resulting in the need to differentiate numericallybetween multiple epochs to obtain the required speeds. This would inevitably lowerthe accuracy of the solution, it would result in the average speed of the time betweenepochs. Secondly the output frequency of the receiver (10 Hz) is lower than frequencyneeded for the radar system (50 Hz).

Both problems might not be the end of it however. Septentrio has already reservedspace in the appropriate message blocks for angular velocities. To obtain the mostaccurate results, the relative speed between the two or three antennas will probablybe used for these messages. This will give results with a theoretical precision of 2.12(mm/s) based on the preliminary calculations in appendix E. This would result in atheoretical standard deviation of the angular velocity of approximately 0.03 (/s). In-terpolation to 50 (Hz) might obtain acceptable results, on the basis that Superbus willbe a high speed vehicle, but will not endure large accelerations. For passenger comfortand safety, these need to be kept to levels comparable to high speed trains, smooth andaptly dampened.

Software

Although GPS data is initially not needed for the hardware in the first Superbus vehi-cle, this does not make high precision GPS obsolete. A range of software can still usethe positioning solutions provided by the receiver. These software applications willnow be explored in more detail.

As an example of software applications vehicle routing will be discussed first. Thiswill clarify the direction the development of the Superbus software may take. Although


no high precision GPS is needed for this application, it may be of importance. Initiallyonly static routing will be used, similar to the simple navigation devices found in mostmodern cars. But remember one of the key philosophies of the Superbus concept: flex-ibility. Superbus will try to cater on an on-demand basis. This means that most routesand driving times are not set in stone and change frequently. Because of this an activeor online routing system may be of great interest.

Envision this scenario as an example. The Superbus is on a trip to Amsterdam andthe normal highway is blocked due to high traffic. Although the Superbus will initiallyexperience no hinder of this on the supertrack, it will need to drive trough the sameaccess routes to the highways as normal traffic. These might be congested as well. Itmay then become useful, to take an alternative route with less traffic on the local roads.

Online routing can aid in this situation to minimise delays. But it will also havethe advantage of estimating arrival times. It will give passengers more detailed infor-mation about delays and will allow them to adapt to them more easily.

Online route planning can in the long run also aid in areas that are less obvious.The Superbus is an electrical vehicle. This has benefits, but also drawbacks. Charginga vehicle takes a lot longer than the refuelling of a conventional bus, several hoursinstead of a few minutes. It is therefore important to optimise the use of the availableenergy. Efficient driving speeds and especially accelerations can extend the range ofthe vehicle significantly. Although this may not always be possible, a system provid-ing online route and traffic information can select more suitable routes depending onrange and remaining charge. This can for example mean a slower or longer route, butone that will provide the Superbus a road on which it can drive at a constant speed,using energy more efficiently.

The next example is an example of a Location Based Service (LBS). A servicethat is able to provide or manipulate information on the basis of the user’s geographi-cal location. The example previous example illustrates this. The intended route of thevehicle can change on the basis of up to date traffic information, and the location ofthe vehicle itself. Usually the information provided is volatile and is therefore unfit tobe stored in a local database. Online databases are more fit as they can be updated fre-quently and accessed by multiple users at the same time. An online database howeverrequires the user to have access to some form of mobile communication. For Superbusthis is the case, providing the Superbus with interesting and important opportunities.

One of these opportunities has already briefly been discussed in section 2.4. There,an online database was suggested to record several parameters concerning road con-ditions, which could be redistributed to the vehicles. Up to 7 radars will be used inthe Superbus in order to gain information of objects around the car. This will help theSuperbus to cope with unforeseen objects on the road ahead at high speeds. This ishowever, the last line of safety precautions. This means that the vehicle must react im-mediately to any event that might be hazardous. This includes braking for wildlife, butalso for unwanted objects on the road. In this last case, it is important that the situationdoes not repeat itself. Emergency braking is dangerous, and at best uncomfortable for


the passengers. Therefore a continuously updated database of the routes the Superbustravels may be desirable. It can store information about the state of the track, locationof bumps and holes, location of foreign objects, or any other information that is neces-sary to keep the vehicle safe. The radars or drivers could make tags for positions whereproblems have arisen. Other vehicles can then be informed with ample time to avoidemergency braking situations or other uncomfortable manoeuvres. If the problem isresolved, this can then be updated in the database. Such a system may be a desiredextra measure of safety. The driver will be offloaded, either because there is ampletime to react, or because non critical information can be suppressed from the driver inthe case a potential hazard is being approached. Also the database will always be up todate. This is a big advantage in contrast with a more conservative on-board version ofthis system. Pondering this for a second can reveal the potential of these types of lo-cation based services, especially if these types of systems are combined and integrated.

In section 2.1 the basic principles of the Superbus system have been explained.Now consider the Supertrack and the online reservation system stem from these prin-ciples for the next example. Take these systems and combine them with the locationbased service approach shown above. A large fleet management system (FMS) thatcombines many of these elements can now be contemplated. Online routing, Super-track states and schedules, passenger reservations and pick up, traffic information canall be combined. An example of such an FMS system is shown in figure 4.8. Thissystem will try to combine all these factors with vehicle locations and schedules andsearch for a more optimum solution compared to a preplanned static route and sched-ule.

Below the system of figure 4.8 will be examined. Note that this is a hypotheticalsystem and by no means complete or final. It serves as an example to demonstrate thebenefits of an LBS approach. Plans for a less elaborate LBS system exist and involveonly a database for road conditions. This example may be used to expand the plannedsystem over time.

This example of the FMS uses information from seven sources to optimise thesolution for three parameters:

• Routing and driving strategies

• Pick up/deliver and vehicle scheduling

• Supertrack scheduling

The seven information sources are:

• Routing and driving strategies

• Pick up and vehicle scheduling

• Supertrack schedule


FMS

Supertrack SchedulingPick up and Vehicle

Scheduling

Routing

And

Driving Strategies

Supertrack State

InformationTraffic Information

Passenger Reservation

System

Superbus

Received Information

· Destination

· Routing information

· Track conditions

· Driving strategies and

Speed limits

· Warnings

· Schedule

Transmitted Information

· Position/velocity/

Integrity information

· Battery information

· Maintanance

information

· General information and

vehicle propertiesOther Superbus

vehicles

Figure 4.8: Superbus FMS example.


• track conditions and traffic information

• Reservation information

• Other Superbus vehicles

The FMS computer is the main component in the system. It will compute a pick upschedule, schedules for the vehicles, routes, Supertrack schedules, and driving strate-gies (maximum speeds, etc.) based on information collected from all the sourcesstated above. The computed information including designated information from theother information sources will then be sent through to the appropriate vehicles. It isimportant to notice the multiple feedback loops present in the figure. The whole sys-tem is dynamic and each of the calculable parameters mentioned above can have theirown optimum. Local situations change continuously, as do reservations and the otherparameters. The FMS will attempt to find a global optimum solution for the wholesystem, with the help of these feedback loops.

Looking at the vehicle side of the system, it is easy to see the benefits. The vehiclesupplies all current and applicable information to the FMS: position information, bat-tery information, but this may also include velocity and GPS integrity data, updates intrack conditions and maintenance information. In return it collects up to date informa-tion: routing, driving strategies, speed limits, scheduling, passenger information, trackinformation, traffic information etc.

GPS integrity data might appear redundant, but consider the vehicle with an activesuspension. If integrity is low, the FMS might decide to lower speed limits, or changethe driving strategy to accommodate the lowered situational awareness. Battery in-formation can be used in conjunction with the passenger reservations to optimise thedriving schedule of each individual vehicle. The real-time nature of the system can beof other interests as well. Information from other Superbusses can be used to quicklyupdate other vehicles. This might be in the area of track and traffic information, butthis may also include maintenance information for example. This could improve downtime of vehicles and increase safety.

Lastly a note on receiver benefits of an LBS system. In the next chapter it willbecome clear that urban environment and multipath are challenging scenarios for theSuperbus Positioning System. Wrong ambiguity resolutions do indeed occur and thereceiver can struggle to adhere to Superbus requirements in these situations. Here aswell, an LBS may be beneficial. As an added measure to minimise wrong ambiguityresolution and multipath, one can in the future investigate the use of a database of visi-bility levels or useful elevation angles on much travelled routes and urban areas. Sucha database is relatively easy to obtain [45]. The database can warn drivers to adverseconditions depending on location, or perhaps eventually, the use of the information canbe integrated in the GPS receiver itself, filtering out or weighting satellites dependingon elevation angle, azimuth and vehicle location.


Current situation, concluding remarks

With GPS -for the moment- being omitted for use in two critical Superbus hardwaresystems, one could argue that the requirements set forth, have been compromised. Thisis not true however. The sections above have shown that the inclusion of centimetrelevel GPS positioning will greatly enhance the flexibility of the Superbus developmentin the future. As the project matures, operational aspects of the vehicle will be bet-ter explored, and a multitude of enhancements can be thought of. Not the least ofwhich, an FMS as was explored in the previous section. Indeed a basic FMS is alreadyplanned in the concept discussed section 2.1. Though initially it will not require thedegree of precision currently offered by the GPS system, it will begin to offer some ofthe aspects discussed above. It will provide an online routing service. Not autonomousas hinted in the previous section, but via a system controller, overlooking the completeSuperbus fleet. Nonetheless in will allow the Superbus to adapt to prevailing condi-tions faster and more easily. The routing also implements the possibility to take thevehicle size into account and exclude certain infrastructural situations.

The planned FMS is basic, but offers a starting point on which extensions, includ-ing the ones discussed in the previous section, can be build. The potential of onlineFMS systems is great and might even evolve to some rudimentary autopilot aspectsthat could regulate the allowed maximum speed of the vehicle, by position. This con-troller could quite easily be implemented, given an accurate, high integrity, positioningsolution and a proper database.

One last aspect which justifies the realisation of the accurate GPS system in theSuperbus is vehicle testing. The vehicle will initially be used for testing purposes,both for the vehicle behaviour itself as well as a testbed for TNO (Toegepast Natuur-wetenschappelijk Onderzoek) research. Vehicle dynamics for example will be studiedin detail. It is of no surprise that an accurate GPS solution is not only valuable butmandatory. With the help of GPS data the vehicle track can be recreated accuratelyand behaviour to exposed situations can be analysed in detail. This will prove veryimportant in order to assure the Superbus is safe not only in intended situations (highspeed, cornering, braking), but also unintended situations such as skidding. The Su-perbus vehicle does not qualify as one of the vehicle categories currently employed byRijksdienst voor het wegverkeer (RDW), hence there is no fixed legislation to whichthe vehicle has to comply. For admission to the open road network, Superbus needsto validate the vehicle is safe for use. An extensive test campaign is necessary for thispurpose, in which the gathered GPS data can play an important role.

Chapter 5

Validation of Superbus PositioningSystem

In the previous chapters, the potential merits and significance of the proposed Super-bus Positioning System have been reviewed. It has become clear that augmented GPSis in theory able to provide Superbus with the necessary data. Now, after acquiringa functional system, it is time to review it in real world conditions, and establish itsperformance and behaviour. This vital step in the development of the Superbus Po-sitioning system is the goal of this chapter. It is famously summarised by a Dutchcomputer scientist: In theory, there is no real difference between theory and practise.But, in practise, there is.

This chapter will start with an insight in the validation process, and certain con-cepts will be clarified. Then, a precise review will be presented of the performancerequirements. The following sections will involve actual testing. First the experimentsare explained and the set up of the tests will become clear. The last section will reviewthe results of the tests and will conclude whether they adhere to the requirements andthe stated specifications.

5.1 Introduction to validation

This section serves as a general introduction to the verification and validation of theperformance of the positioning and heading system for the Superbus. The systemwill now be scrutinised under various conditions in order to obtain field data. Withthis test data available, the system will then undergo the verification and validationprocess. Definitions of these concepts are given below. Validation and verification willbe covered mainly in section 5.4.

• Verification: The task of determining whether a system performs to its specifi-cations.

• Validation: The process of determining whether a system fulfils the purpose forwhich is was intended.

79

80 CHAPTER 5. VALIDATION OF SUPERBUS POSITIONING SYSTEM

To be able to review the performance of the GPS system with the processes dis-cussed, parameters must be established in order to quantify the positioning perfor-mance. Yet there is no standard present for this type of vehicle, or GPS system. How-ever, since the Superbus clearly falls in the category of professional use, its GPS systemwill be reviewed with an adapted version of the aviation performance parameters [1].These parameters are, by using stringent requirements, used to verify navigation so-lutions, and establish if a system is fit for use in safety-critical applications, such asaviation and maritime navigation.

For the system as installed in the Superbus, performance parameters are cate-gorised into three main components: Accuracy, integrity and availability [16, 36]. Thefourth component used in aviation, continuity, is omitted in the Superbus assessment.Continuity involves the ability of a navigation solution to complete an action, such as alanding, without accuracy or integrity problems. The navigation in the Superbus needsto be continuous, hence there is no specific action to complete. The remaining threeare discussed in more detail below.

Accuracy: Measure of navigation output deviation from truth, usually expressed as1σ or 95% (approximately 2σ) error limits.

In less technical terms: How close is the calculated position solution to the trueposition? An example. As was previously addressed in chapter 3, the accuracy of astandard stand alone GPS receiver is usually several meters in favourable conditions.The estimated position can be several meters away from the antenna with which youare measuring.

This has consequences for the application of the system. If the required positionaccuracy is on the meter level, a position error of several centimetres, is of lesser im-portance. And indeed, a system which provides a better accuracy will most likely beunnecessarily large and expensive. Hence care should be taken into properly assessingthe requirements of the position accuracy as well as other performance parameters.

There is however more to say about the subject. The definition noted above is takenfrom ICAO annexes, but is actually not the complete story. Notably there is a differ-ence between accuracy and precision, which is hidden in the definition above. Bothare distinct and will be addressed separately in section 5.4. Precision is defined as thedeviation of a single solution from the mean of the complete data set. The summationof all these solutions results in a probability density function for the complete data set.The mean of this set is however not necessarily the actual or ”true” mean. This is whywe also introduce accuracy, which involves the mean of the complete data set and howmuch it deviates from the truth, see also figure 5.1. The accuracy of a data point canbe defined as the precision plus a bias, as systematic error. An example: the positionestimates of a data set are all within a few centimetres from each other. However, thecomplete data set, is several meters away from the target, or in GPS terms, the groundtruth. In this scenario the precision of the data set is very high, the accuracy on theother hand isn’t.

5.1. INTRODUCTION TO VALIDATION 81

Figure 5.1: Bias and precision graphically represented, accuracy = precision + bias.[source: wikipedia]

The results in section 5.4 will discuss accuracy and precision separately. We willalso discuss bias: the difference between the mean of the measurements and the refer-ence value, or ground truth. The precision of the GPS system will be quantified as the1σ values, or empirical standard deviation of the data set. This is represented in equa-tion 5.1, where N is the number of measurements, xi the solution of a measurementand x the mean of the data set.

σ =

√1N

n

∑i=1

(xi− x)2 (5.1)

Note that the standard deviation values will be listed with intervals. Since theobtained data sets are empirical, the results might be slightly different, if the test is re-peated. In other words, the standard deviation has its own standard deviation, σσ. Forthis reason the standard deviation is given accompanied by the extreme values betweenwhich the standard deviation will lie with a 95% confidence level.

The last item listed, will be the 95th percentiles. These percentiles indicate thevalue below which, 95% of the recorded solutions can be found. The reference valuefor this figure is the ground truth, and hence it is a measure of accuracy.

Integrity: A measure of trust which can be placed in the correctness of the informa-tion supplied by the total system. Integrity includes the ability of a system to providetimely warnings when the system should not be used for it intended operation.

For safety-critical applications in practise this means that a system should be ableto autonomously assess whether it exceeds specified tolerance levels, and act accord-ingly. An operator requires a warning when the accuracy of the estimated position


solution from the receiver is not within a certain guaranteed limit, due to anomalousmeasurements (outliers, etc). For this reason the Septentrio receiver used in the Super-bus GPS-system, makes use of Receiver Autonomous Integrity Monitoring (RAIM).This system uses statistical tests to evaluate the integrity of the computed position so-lution. It uses surplus measurements of GPS satellites, to assess the consistency of themeasurements. To be able to do this, certain quantifiable performance requirementsare needed: maximum probability of missed detections (Pmd), maximum probabilityof false alarms (P f a) and alert limits.

As can be seen in figure 5.2 there are four possible outcomes when detecting out-liers in the measurements (e.g. due to a failing satellite, a malfunctioning receiverchannel, or multipath). Two are correct: a detected outlier, when an outlier is present(D), and the inverse, no outlier is detected, when no outlier is present (B). Two out-comes on the other hand are erroneous: a false alarm due to outlier detection whenno outlier is present (A), or an undetected error, no outlier detected when an outlieris present (C). The statistical tests are based on assumptions that the residuals of thesatellite measurements, the difference between the measured and estimated distance toa satellite, are normally distributed, and the mean is equal to zero for all correct mea-surements (blue curve, in figure 5.2). When, for example, a satellite is not performingproperly or multipath is present, the residuals will form a distribution with a mean un-equal to zero (red curve, in figure 5.2). Naturally these two distributions will intersect.allowing the occurrence of missed detections (C).

Figure 5.2: Outlier detection. [source: Septentrio]

The procedure to monitor integrity used in the Septentrio receiver is based on atest procedure developed at the TU Delft [34, 48] and follows three steps:

• Detection: Assess whether some anomalous behaviour has taken place.

• Identification: Assess which measurement constitutes as an anomaly.

• Adaptation: Remove measurement from position computation to restore in-tegrity.

5.1. INTRODUCTION TO VALIDATION 83

To detect whether an (unspecified) anomaly or outlier has occurred, an overallmodel test is used. If this overall model test statistic is larger than the predeterminedthreshold of the probability of a false alarm (A), the procedure performs a statisticaltest (w-test) to evaluate from what satellite the detected outlier is coming from. Thissatellite is then removed from the position computation and a new solution is com-puted.

To evaluate the impact of missed outlier detections, the minimal detectable biases(MDB) of the satellite pseudoranges are computed: The MDB’s show the minimumrange error for each satellite pseudorange measurement that can be discerned by thew-test described above with the predefined Pmd. The receiver can then calculate theimpact that these MDB’s have on the position estimate, resulting in reliability levels(RL’s). In other words: suppose an undetected outlier introduces an error in the rangemeasurement of a satellite, with the size of the MDB. The effect that this undetectedoutlier has on the position estimate, can be compared with the estimate where this out-lier is present. The results of this comparison are the reliability levels (RL’s), or errorsin position estimates which cannot be detected by this integrity monitoring. RL’s thusgive a radius within which the accuracy of the position is ensured. The user then eval-uates whether or not, this reliability level is sufficient to use the GPS system. This iscommonly done by setting horizontal and vertical alert limits (HAL and VAL). Whenthe horizontal or vertical RL’s exceed this limit, the calculated position should not beused for navigation purposes.

As becomes clear from the above, the calculation of the integrity of a GPS systemis rather complex and also dependent on user requirements: the maximum probabilityof missed detections, maximum probability of false alarms and alert limits. To ensureintegrity it is essential that these values are properly defined. Only then it becomesclear whether or not the position estimates meet the requirements and may be used forcritical applications.

Availability: Fraction of the time the navigation system is usable (as determined bycompliance with accuracy and integrity requirements).

Availability is the percentage of continuous time, during which the system is ableto deliver position solutions that fall within certain requirements of accuracy and in-tegrity. These are often HAL and VAL values determined by the user. For example,consider a certain GPS system, which is able to deliver centimetre accurate positionsolutions for 6 hours every day, and meter accuracy during the remaining 18. If theaccuracy requirement is 10 centimetres, the availability of the system will only be25%. If however, the requirement is 10 meters, the availability jumps to 100%. Sothe concept of availability is not only dependent on the space segment, but also on therequirements of the user.

In this thesis, Availability will not be assessed in great detail, only the observed re-sults will be specified and commented upon. Statistically significant availability testingwould entail controlled long duration testing, both static and dynamic. This was not


done, and would for now be of lesser importance, due to the tendency of UMTS dis-connections during dynamic testing, currently clearly the weakest link in the system.This is largely uncontrollable in real world testing, and dependent on prevailing con-ditions of the mandatory 3G network. This is dependent on geography and serviceprovider. One might argue that availability entails the whole system and hence alsothe availability of the 3G network. This is correct and therefore the values are dis-cussed in section 5.4. However UMTS service is, on the moment of writing, still beingrolled out, the UMTS network will only expand, and national coverage, and thereforeavailability, will only become important in later stages of Superbus development.

5.2 Performance requirements

For Superbus accuracy and availability requirements were set to ”best possible solu-tion attainable”. This has been so from the start of the project, and is partly due tothe nature of the Superbus project. The vehicle is a prototype, an attempt to launcha new way of thinking in public transport, and as a result a high profile technologydemonstrator. The nature of the project is to be on the very top end of current technol-ogy as the project progresses into maturity. ”Best possible”, of course, has its own setof boundary conditions as funds and time frame are limited, immediately narrowingthe positioning solution options considerably. In conjunction with the MathematicalGeodesy and Positioning group (MGP) a base requirement was therefore set: 95% ofthe obtained solutions should fall within 5(cm) of the ground truth.

But even without the direct need of the requirements set conjointly by Superbusand MGP, the system can still be evaluated to it merits. The system can be verifiedby comparing the actual system performance to the performance figures given by themanufacturer. Septentrio has provided these figures and they are presented in table 5.1and table 5.2. The first represents the heading and pitch precision for different baselinelengths. The second table shows the horizontal and vertical precision. Note that forthe RTK solution, the precision is supplemented with another value in part per mil-lion (ppm). This signifies that an RTK solutions degrades in precision, with increasingdistances from the base station. For the horizontal solution 1 part per million, or 1cm every million centimetres, or 10 (km). The full specifications of the PolaRx2eHreceiver are available in appendix A. These provided values, along with the require-ments set by Superbus, will be used in the verification and validation of the receiverperformance.

The requirements for integrity are more difficult to define. These requirements arelargely determined by the accuracy requirements. Notifications must be given, whenthe system can not be guaranteed to adhere to certain performance values. However,the vehicle is a technology demonstrator, and no legislation exists stating the integrityrequirements of a positioning system used as a primary navigation system. For avia-tion application, legislations and requirements are well documented [11, 23], and needto be met, before any system is certified for primary navigation. Such legislation is notin place for road vehicles, and indeed the whole Superbus vehicle itself falls outsideany category road vehicle as defined by the dutch government.

5.2. PERFORMANCE REQUIREMENTS 85

Table 5.1: Heading and pitch precision (1σ) specification, provided by receiver manu-facturer.

baseline length heading precision pitch precision(m) (deg) (deg)

1 0.30 0.603 0.10 0.2010 0.03 0.06

Table 5.2: Positioning precision parameters (1σ) specification, provided by receivermanufacturer.

precision horizontal vertical(m) (m)

Stand Alone 1.1 1.9RTK 0.01 + 1ppm 0.02 + 2ppm

For now the Superbus vehicle runs as a test vehicle, and legislation procedures toaccommodate such an unusual vehicle as the Superbus are underway. It will becomethe task of the government to determine HAL, VAL and failure rate values, acceptablefor critical GPS use on the public road. For this reason, no HAL and VAL values aredetermined as of yet, nor any failure rates. The integrity monitoring, RAIM, integratedwithin the receiver can, at any point in time, be adapted to future requirements. Eval-uation of the RAIM system and its availability is however outside the scope of thisthesis and has been done extensively [16]. Therefore the RAIM system is, for purposeof this rapport, deemed fit for use in the Superbus.

However an indication of the necessary integrity requirements can be suggested,originating from the aforementioned aviation requirements. First of all it is importantfor Superbus to assess what the critical systems are that will use GPS, and what therequirements for these systems are. For example, in the case of active suspension (sec-tion 4.5), the requirements for GPS, counter intuitively, may actually be less stringentat high speed than at low speed. Since the active suspension idea depends on timing,a one meter error will only cause a 0.014 second timing error at 250 km/h, while at50 km/h it will be 5 times as high. This illustrates the importance of a precise set ofrequirements. Once these are available, one can start to set the HAL and VAL valuesin the RAIM algorithm, perhaps as a function of speed. The maximum probability ofmissed detection can as a start be set identical to the recommendation for a CAT-I ILSlanding at 10-7.

In conclusion the test results will be used for verification and validation of theSuperbus navigation system. Firstly the compliance to Superbus requirements will beassessed. Furthermore the performance claims of the receiver will be verified, moreprecisely for position and heading precision. Lastly the results section will comment


on the availability of the system.

5.3 Experiments

In the previous section it has become clear what aspects of the receiver need to beevaluated. By using three main performance parameters, accuracy, integrity and avail-ability, we are able to make qualitative remarks about the performance of the SeptentrioPolaRx2eH receiver, in conjunction with the complete GPS system, as developed in4.3.

A crucial aspect of performance testing is a reference, a ”true” position, to whichto compare the real time solutions output by the receiver. Only by comparing the re-ceiver to this ground truth are we able obtain figures regarding the performance of thereceiver. In practise it is impossible to obtain a perfect ground truth, since this groundtruth is obtained by measurements as well. We must therefore suffice with the ”bestpossible” ground truth that can be obtained, depending on test set up and conditions.This aspect of testing will be discussed in more detail in the following sections.

The data for this performance evaluation was gathered by three test campaigns,two static and one kinematic, consisting of several individual experiments. The goalsof each experiment was different, or at least overlapping, and together provide enoughdata to assess the quality of the receiver and its navigation output. First a general set-upand the goals of each is provided. Then the technical set-up of each campaign will begiven, to provide a better understanding to how the data was gathered. Furthermore,general remarks will be presented which are important during the processing of theresults. Although the kinematic test was performed prior to the static test, the statictest will be reported first as the kinematic results will be discussed with reference tothe static results.

5.3. EXPERIMENTS 87

Materials testing

In December 2007, a general equipment test was performed. This test was combinedwith a materials test. In section 2.5 it has briefly been suggested that the placement ofthe two GPS antennas is quite important. This is, amongst others, supported by table5.1, where the manufacturer supplied heading and pitch accuracies are represented.One can see that the accuracy is proportional to the baseline length.

Yet the Superbus design is relevant as well, and provides it’s own consequencesto antenna placement. Superbus is a meticulously shaped vehicle, designed in orderto reduce drag to a practical minimum. Therefore, on the outside of the vehicle, noobstructions in the airflow were to be created. Since the (already) acquired anten-nas are relatively large professional grade antennas, this meant that they were to beinstalled internally. This has several consequences. Antenna placement flexibility isreduced significantly due to shape, size and placement of the internal bays of the vehi-cle. These dictate where and how the antenna’s can be placed.

More importantly risk of obstructed aerial view and multipath are increased. Themain problem is that the internal bays wherein the antennas are installed are coveredwith sheet material in order to satisfy aerodynamic and esthetic demands. The antennawill therefore be covered at all times, when the vehicle (and thus the GPS system) is inoperation. It is therefore of importance to have an apprehension of the effects the usedmaterial will have on the GPS signals the antenna is trying to receive. Some materialsreflect or absorb Radio Frequency signals. This behaviour could, in this case, prohibitthe satellite signals to arrive at the antenna that is covered and hence make the com-plete GPS system useless.

Since the antennas would always be covered, and the consequences possibly verylarge, it was decided that a materials test would be conducted. To cover the antennasin the available antenna bays, two material possibilities were available.

The first and original possibility was a 2 millimetre thick carbon fibre sheet. Sincecarbon is a conductive material as well, it would potentially prohibit radio signals toreach the antennas in much the same way as steel. The performance is dependent onthe thickness and the weave of the carbon sheet.

The second possibility was an equally thick advanced thermoplastic: High Perfor-mance thermoPlastic Composite (HPPC). Since this material was quite new, the exacteffect on radio signals was unknown. The material is based on a thermo-moldableresin reinforced with a glass fibre filler. Glass fibre is normally non conductive, andwill usually not interfere with radio signals. Therefore it is potentially a better can-didate for the cover sheet. However, the absence of exact material specifications stilldemanded further testing.


Experimental set-up materials test

The test took place on December 4th 2007 in a field near the faculty of Aerospaceengineering in Delft (see figure 5.3). The field was selected due to its proximity to thefaculty, but more importantly because of the unobstructed aerial view the low multi-path environment. This will help the solution to be as accurate as possible, and elimi-nate these factors from the obtained result as much as possible. The day was overcastand there was a slight drizzle, the afternoon test period was centred around a windowwhere no less than 11 satellites would be visible at any one time. This is demonstratedin figure 5.4.

Figure 5.3: Location of the material test. [source: Google Earth]

Figure 5.4: Number of satellites visible during materials test (cut-off angle 5).

5.3. EXPERIMENTS 89

The material set-up was as follows. The base station, the PolaRx2e, was set upwith AeroAntenna AT-2775 antenna on a tripod and allowed to measure undisturbedfor the complete test. The receiver was set to default values, but set to track only GPSsatellites with a cut off angle of 5 degrees, and record the resulting Rinex files.

The other receiver, the PolaRx2eH, was equipped with the same antenna, and wasset up approximately 10 metres away from the base station. This receiver was set-upto approach the Superbus settings. These are available in table 4.1. The exception tothese settings is the cut-off angle, which was set to the same 5 degrees as the basestation and the output rate, set to 1 Hz. Again the receiver was allowed to log Rinexfiles.

The experiment itself was set up as follows. The Base antenna was allowed to logthe raw measurements for the complete duration of the experiment. The measurementswere then processed in Trimble geomatic office version 1.63 to obtain a static solution.These coordinates consequently served as the base station coordinates. Subsequentlya baseline could be set up between this receiver and the secondary, or rover receiver.

The rover receiver was also allowed to log during the same time period withoutphysical disturbance. However each of the two selected materials would each coverthe antenna for approximately 30 minutes. A construction of 3 wooden poles wouldensure that the material sheets would completely cover the antenna, but not come intocontact with it. The construction can be seen in figure 5.5. The first 30 minutes of rawmeasurements where recorded without any material, though the wooden poles werealready present. These measurements would provide a base solution, to which theperformance of the two materials could be compared. For the second 30 minute timeframe the rover antenna was covered with the HPPC material. The receiver did notmove, nor did it stop logging the raw measurements. This procedure was repeated forthe third 30 minute time period with the carbon fibre material.

To obtain a solution, a baseline was set up in Trimble geomatic office, between thebase station and the rover. This baseline was computed both statically and dynamically.This was done for all three measurement periods. In this way the static solution can becompared to the remaining two solutions. This is done in section 5.4.

Static testing

The static testing was done on march 31st 2009, on the rooftop of the NMI (Ned-erlands Meet Instituut) building in Delft. This location was chosen because of theavailability of a highly accurate ground truth. The building is a stable platform. Themain column on which the testing is done, is constructed upon its own separate foun-dation, free from the remainder of the building. This allows for steady measurementsfor longer periods of time and therefore a more accurate ground truth. Position accu-racy of the antenna mount points, or markers, are within a single millimetre from eachother (see http://gnss1.tudelft.nl/dpga/station/Delft.html), and therefore very suitablefor static testing.


Figure 5.5: Rover antenna covered with carbon fibre material,supported by the woodenconstruction.

In figure 5.7 a schematic overview of the roof is given. As can be seen, severalfixed points are marked, and represent markers on the building of which the positionsare precisely known. By setting up a base station on point 21, and configuring theSuperbus segment on points 13 and 14, a suitable platform for static testing is created.Though not identical in set-up as the Superbus itself, the Superbus set-up, as describedin 4.3, will confidently provide equal results in similar conditions.

The goal of the static tests performed, was to evaluate the static behaviour of theGPS system in normal circumstances. Furthermore, after initial kinematic testing, itbecame clear that more data was needed to assess the reacquisition behaviour of thereceiver. Therefore two tests were scheduled. The first was a standard static test, wherethe receiver was set-up in the way it will initially be set up in the Superbus, see table4.1. Approximately 30 minutes of data was collected and stored.

5.3. EXPERIMENTS 91

Figure 5.6: NMIbuilding and test set up.

Figure 5.7: Overview of the top of the NMI building.

The second test required loss of lock situations, encountered by the Superbus forexample by passing under a bridge, or travelling through a tunnel. This needed tobe simulated repeatedly and under controlled conditions, to ensure clean and reliabledata. The need for a controlled situation, had emerged during a previous kinematictest, where it proved difficult to obtain enough data under similar condition, to reliablyformulate propositions regarding the reacquisition time of the GPS system. This re-sulted in a static experiment where both receiver antennas where covered repeatedlyby metal objects to simulate loss situations. Both the main antenna on point 13, and theauxiliary antenna on point 14, where covered for 5 seconds, ensuring a proper signalloss for all satellites and both antennas. Then the receiver was allowed to regain a RTKfixed solution, before repeating the process. This was done a total of 56 times.

Experimental set-up static test

As can be seen in the figure 5.7, placing the main antenna of the Superbus segmenton point 13, and the auxiliary antenna on point 14 yields the characteristics presentedin table 5.3. The antennas used for the Superbus segment were Leica AT502, whilefor the base station a Leica AT504 without radome was used. Because two antennas


of differing types have unequal phase centres, this must be corrected during post pro-cessing. The test set up will be largely the same as Superbus setup, it will howeverdeviate from the actual Superbus configuration in 4 areas: utilised antenna, loggingspeed, transmission of the correction data and base station location.

Firstly, the antennas used for the Superbus segment were two Leica AT502 anten-nas, while for the base station a Leica AT504 without radome was used. These are notthree identical AT2775 antennas (rebranded to PolaNt), as is the case in the Superbus.The use of different antennas can introduce a slight difference in performance. Thesedifferences will be small though, due to the fairly similar performance of both antennatypes.Secondly, because these base station and rover antennas differ, they will also have un-equal phase centres, which has to be corrected during post processing.

Table 5.3: Antenna positions and baseline lengths.

Type Antenna Type Marker Baseline Baseline length (m)

Base station Leica AT504 point 21Point 21 to 13 14.984

Main antenna Leica AT502 point 13Point 13 to 14 3.002

Auxiliary antenna Leica AT502 point 14

The phase of the centre of a GPS antenna is not a single point [24]. It varies,depending on where the satellite signal comes from. Given these variations, usingdifferent antennas for a single test setup is undesirable. The unequal antennas andconsequently differing phase centres will typically introduce a vertical error that canbe difficult to handle. Nonetheless, point 21, with its AT504 antenna, was still selectedfor use as a base station. This was due to renovations on the NMI building, whichprohibited certain markers from being utilised; hence no suitable point was availablefor a third AT502 antenna as a base station. Point 21, therefore, provided the bestsignal, with manageable multipath. The vertical separation of the phase centres issubsequently handled in post processing by incorporating the calibrated phase centrefor the two different antenna types, and adding the resulting difference to the antennaheight of the base station. These calibrated phase centres are acquired by and availablefrom the United states National Geodetic Survey (NGS). Note that the phase centresprovided by the NGS are only in the vertical direction, caused by satellite elevation.Azimuthal changes are neglected as they are caused by the local antenna environment,and are therefore hard to correct.

The receiver settings were nearly identical to those anticipated for the Superbus,available in table 4.1. This permits a direct comparison to the results that can beexpected for the vehicle in a static position. The single exception is that the outputinterval was set to 1(Hz), in favour of 10(Hz), due to logging constraints.The proximity of the base station, and the resulting short baseline, is not expected tointerfere with this comparison, but should be noted. The operational Superbus can usevirtual base stations, as was revealed in section 3.4 which will typically mimic base

5.3. EXPERIMENTS 93

stations at a distance of 5 to 10 kilometres. The results obtained here should thereforebe slightly better (approximately 1 centimetre) than can be expected for the Superbus.

The correction data, needed for an RTK solution, consisted of RTCM 2.3 messagessent every second. The messages where transferred directly from the base receiver tothe rover receiver, using a RS232 com cable and a null modem. Note that UMTS wasnot utilised during this test, as it was already derived from testing that the system asseen in section 4.3 works well under normal circumstances. Lastly note that the aux-iliary antenna is only utilised to obtain heading and pitch values, the position solutionoriginates from the main antenna.

The loss of lock for the reacquisition test was obtained by blocking both antennasfor 5 seconds by hovering a frying pan over them. The receivers were then allowedto re-establish a RTK fixed position and heading solution. The blocking interval wasselected to be 30 seconds, meaning that if the re-acquisition time, after the initial 5 sec-ond signal blockage, was 20 seconds, 5 seconds of RTK fixed data would be recorded.If the reacquisition time exceeded the 30 second mark, the next 30 second intervalwould be used completely before instigating the succeeding loss of lock.

Kinematic testing

Kinematic testing was performed on the 23rd of May 2008, and involved a vehicleequipped with copious, near excessive amounts of GPS receivers. The testing for theSuperbus system was part of a larger experiment, involving other professional gradereceivers. This allowed the vehicle path to be reconstructed in the greatest possibleprecision, while still maintaining a fully functional vehicle in real world conditions,relatively unrestricted in motion.

The vehicle was a standard, road legal, van equipped with a carefully measuredplatform, on which the receivers where mounted. In figure 5.8 an overview is givenof the platform, and the receivers mounted relevant for the Superbus testing. The ex-periment was generally set-up in the following manner. The Septentrio receiver, wasset-up in RTK rover mode, with settings matching those used in the Superbus scenario.RTK corrections from a base station on the NMI building were collected in the vehicleup to 7 kilometres away, via UMTS receiver. Corrections were transferred using theNTRIP protocol, as explained in section 4.2. The receiver was allowed to acquire itsown real time solutions, and were only logged for analysis. The two other antennason the platform were connected to two industrial grade Trimble R7 receivers, loggingmeasurements. These measurements could then be post-processed to a ambiguity fixedcarrier phase solution using a reference antenna situated on the NMI building. This al-lows the vehicle path to be accurately recreated. Values obtained from the Trimblereceivers where used as a ground truth for the real-time solutions of the Septentrio re-ceivers. The ambiguity fixed carrier phase ground truth was the best possible referencethat could be obtained in this real time kinematic experiment, while still maintaininga relatively free roaming vehicle. The Trimble track is hence not perfect, with a stan-dard deviation of a few centimetres, depending on conditions and distance to the base


2.1

95

m

2.7

95

m

0.700 m

0.693 m

0.2

91

0 m

0.2

99

4 m

2.59 m

2.60 m

Trimble A

Trimble B

PolaRx2 (1)

PolaRx2 (2)

Dire

ctio

n o

f tr

ave

l

Figure 5.8: Vehicle frame overview, dimensions and antenna placement. Vehicle frontis top of figure.

5.3. EXPERIMENTS 95

station. For the results the post processed Trimble track was however assumed as theground truth.

The goal of the campaign was to obtain data to be able to comment on the realworld performance of the Superbus. A total of three experiments where carried outthat day that will henceforth be known as: (1) Duifpolder experiment, (2) A4 experi-ment, (3) Emerald experiment.

The Duifpolder experiment involved repeatedly driving along a long straight roadin a flat, open countryside environment. Almost no objects were present, which couldblock or reflect GPS signals (see figure 5.23). Several round trips were made, at differ-ent speeds. The main goal was to obtain data under optimal circumstances, however,due to the lack of UMTS reception, a stable connection could not be achieved. Al-though some runs where successful with an RTK fixed solution, the bulk of the datacontained only a stand alone solution. Therefore, in this test the main focus will lie inthe stand alone performance of the GPS receiver. The baseline length of the experi-ment was approximately between 6 and 7 kilometres.

The A4 experiment took place on the final stretch of the A4 highway near Delft-Zuid. The goal here was to see how the receiver behaved in sub-optimal conditions,and to evaluate the reacquisition behaviour of the receiver. The section of highwaycontained 2 viaducts, causing a short loss of lock for all satellites. The receiver thenreinitialises a lock, subsequently a stand alone position, RTK float solution and lastlyan RTK fixed solution. Since the Superbus will drive on, or next to normal highways,it is imperative to know the behaviour of the receiver in these conditions. It will hap-pen on a regular basis that an object blocks satellite signals, causing a loss of RTK oreven a loss of stand alone solution. The data link conditions of the second experimentwere better than the Duifpolder, although problems with the UMTS reception werestill present. The baseline length of this experiment was approximately between 4 and5 kilometres.

The Emerald experiment was a test in light to medium urban environment as canbe seen in figure 5.29. It is a track through suburbs which diminish the view of the sky.Due to the more urban environment, the UMTS reception improved dramatically, andin no instance was the connection for the correction data lost or dropped. This allowsfor a proper evaluation of performance when the GPS system is working as it should,in a real world environment. Please note though that this were by no means perfectconditions, and results will show that multipath has played a role in positioning pre-cision. In addition, most of the time the receiver-set cut-off angle of 10 degrees, wasexceeded by buildings and other infrastructure, diminishing the amount of satellitesvisible. This scenario might be seen as a absolute worst case scenario for a highwaysituation, the scenario where precise positioning is most important. It might be seen asa simulation of an environment the Superbus might encounter on urban highways, forexample near The Hague or Amsterdam. The baseline length of this experiment wasapproximately between 1.5 and 2 kilometres.


Experimental set-up kinematic test

As was discussed earlier, the experimental set-up involved installing a platform on topof the test vehicle. This platform was carefully measured, and the position of eachantenna documented, as shown in figure 5.8. For the purposes of the experiment, itwas assumed that the platform was rigid. This is a reasonable assumption, due to thethickness of the beams, used in the platform, the way it was connected to the roof,and visual assessment during the experiment. The equipment on board relevant for theSuperbus testing included two laptops, two Trimble R7 receivers, two Zephyr geodeticantennas, a Septentrio PolaRx2eH receiver and two AeroAntenna AT-2775 antennas.

Figure 5.9: Vehicle with installed frame.

The experiment was set up as follows. The raw measurements of the two Trim-ble antennas, A and B, where logged by the two R7 receivers every second. Thesemeasurements were later post processed using a reference station, located on the NMIbuilding, resulting in two individual sets ambiguity fixed carrier phase positioning so-lutions, one for Trimble A and one for Trimble B. These sets were calculated usingTrimble Business Centre version 1.12.

With the platform of the antennas assumed rigid, carefully measured and locallevel at all time, it is possible to obtain a ground truth for the main Septentrio antenna.For this, the Trimble B track was translated and rotated for every epoch in the test dataset, in order to obtain a virtual Septentrio ground truth. The translation was equal tothe measurements given in figure 5.8. The rotations where obtained by comparing theTrimble A and B solutions and so deducing the heading between both receivers. Sincethe relation between the heading of the Trimble receivers, and the main Septentrio an-tenna was fixed, it is now easy to calculate the virtual Septentrio ground truth for everyepoch in the data set. This ground truth is used to as a reference in order to assess theperformance of the real time Septentrio solution in all the dynamic test described inthe previous section.

The Septentrio receiver in the test vehicle, received correction data in RTCM 2.3

5.3. EXPERIMENTS 97

from a PolaRx2e receiver, the Superbus base station receiver, which was installed onthe NMI building as well, several kilometres away. The data transfer was conductedwith the NTRIP protocol as explained in 4.2. The UMTS receiver in the test vehiclewas connected to a laptop running an NTRIP client. This client collected the correc-tion data and rerouted it via RS232 cable, to the com 2 port of the receiver. This wasnow able to calculate an RTK solution, logging it at 10 (Hz), both on an internal loggerand on a secondary laptop, running a basic program, designed for use in the Superbus.Once again the Septentrio receiver was set analogous to Superbus settings, keepingthe test variables down to a minimum. The settings of the PolaRx2eH are available intable 4.1.


5.4 Test results

In the following, the results of the staged experiments will be transcribed. Complianceto requirements and manufacturer claims will be verified, a concise evaluation will begiven and a conclusion will be formulated.

Materials test

The goal of this experiment is to judge the effects of a sheet of material covering theantenna, as discussed in section 5.3. Two materials are to be compared: HPPC andcarbon fibre. Three baseline solutions were computed, both static and dynamic. One30 minute period without material covering the antenna, one with HPPC covering theantenna and one with carbon fibre.

As mentioned the Rinex files from resulting from the experiment were processedin Trimble geomatic office version 1.63. The results consisted of three static solutionsfor the rover antenna, and three sets of dynamic solutions, consisting of 1771, 1741and 640 solutions for the no-plate, HPPC and carbon fibre periods respectively. A sky-plot of Delft during the experiment is presented in figure 5.10.

Figure 5.10: Skyplot over Delft, during the materials experiment.

The results of the materials experiment will be presented in relation to the no-platestatic solution. In other words, the 30 minute no-plate static solution will be consid-ered the ground truth of the rover station. All other solutions will be given in relationto this point.

Table 5.4 and figure 5.11 shows a comparison of the 3 static solutions. One canclearly see that the HPPC material has no significant effect on the static solution, boththe no-plate and the HPPC solution are nearly identical. The carbon fibre solutionhowever, differs significantly, showing that indeed carbon fibre is unfit to be used as atop sheet material. The carbon fibre static position is calculated a full 3.70 metres fromthe no-plate static ground truth, while the HPPC Static solution differs no more than

5.4. TEST RESULTS 99

2 millimetres. Note that figure 5.11 only shows the position errors in the horizontalplane.

Table 5.4: Static solution performance results.

Position error from ground truth

HPPC Carbon fibre

North position error (m) 0 2.063East position error (m) 0 0.766Up position error (m) 0.002 2.973

Figure 5.11: Comparison of three, 30 minute, static solutions.

The dynamic solutions predominantly show the same scenario. Figure 5.12a and5.12b show the point clouds of the dynamic no-plate and HPPC solutions. Once again,these show nearly identical behaviour. This is supported by the standard deviationsand 95th percentile values presented in tables 5.5 and 5.6. These values nigh on cor-respond. This is also supported by a two sample F-test of the two data sets. The testcannot reject the hypothesis that both sets have the same variance at a 5% significancelevel, for all directions. Please note that the output resolution of the Septentrio receiveris limited to 1 millimetre.


The carbon fibre figures however are again substantially worse, and it is clearlyvisible that the dynamic solution is influenced by the carbon sheet (figure 5.12c, table5.5 and 5.6). It is also important to note that only 640 position solutions could be cal-culated of the approximately 1800 measured epochs. This means that only about 38%of the measurements was good enough to obtain a solution, but generally still far offfrom the ground truth.

Table 5.5: Dynamic solution performance results, standard deviations.

Standard Deviation

North (m) East (m) up (m)

No-plate 0.0015 0.0023 0.0046HPPC 0.0016 0.0024 0.0046Carbon fibre 1.4485 2.2025 2.2854

Table 5.6: Dynamic solution performance results, 95th percentiles.

95th percentile

North (m) East (m) up (m)

No-plate 0.0030 0.0040 0.0110HPPC 0.0030 0.0040 0.0120Carbon fibre 2.3475 1.6210 2.5580

As a last illustration of the unfitness of the carbon fibre top sheet, two figures arepresented for two individual satellites, visible during the complete experiment. Thefirst figure, figure 5.13a, shows the carrier to noise ratio of a high elevation satellite.The second, figure 5.13b, shows the ratio for a descending satellite. The red verticallines show the three measurement intervals, no-plate, HPPC and carbon fibre. Onecan clearly see that the behaviour of the carrier to noise ratio does not alter when theHPPC sheet is installed over the antenna. The ratio however, falls dramatically, whenthe carbon fibre plate is installed. Although some peaks can still be observed duringthe carbon fibre intervals, the signals remain very weak, furthermore, the peaks canprobably be attributed to multipath.

As a conclusion it can be stated that the use of the carbon fibre cover plates isnot recommended for use on bays in which GPS antennas are placed. The carbonfibre sheet degrades or blocks the satellite signal to such an extent, that positioning isimpossible, or not adhering to the accuracy requirements set in section 5.2. The sheetmaterial would render the GPS system unusable for any practical application in theSuperbus. The HPPC material performs much better, and performance results showsalmost no deterioration when the antenna is covered by the HPPC sheet. Therefore theuse of the HPPC material would be recommended over the carbon fibre top sheet. Asa secondary recommendation, after observing the effects of the carbon fibre plate, it


is advised that proper static GPS performance testing should be done when the GPSsystem is installed in the vehicle. This to evaluate the effects of the carbon fibre bays,in which the antennas are installed. These can still inhibit the system to adhere to theperformance requirements due to deteriorated signal reception or the high multipathenvironment (due to the reflection properties of the carbon fibre) within the bays.

Static test, performance

The goal of this experiment is to assess the static performance parameters of the GPSsystem, as described in 5.3. The parameters relevant for GPS performance are ex-plained in section 5.1, the experimental set-up in section 5.3. Conditions were verygood with a normal amount of satellites visible, see figure 5.14

The data consisted of almost 25 minutes of position, heading and pitch solutions,recorded every second. RTK was available and maintained during the complete testof 1490 RTK epochs, resulting in an unsurprising 100% availability. Do note thatUMTS, and NTRIP where not used. The 1490 epochs were converted and comparedto the known values of the designated points on the NMI building. These known valueswere gathered in a 1997 measuring campaign. Final results of the raw position data arepresented in ETRS89, however relative deviations from the ground truth will be pre-sented here. Also take care in noticing that the results are given in x,y,z coordinatesand not in north, east and up.

In the static test the phase difference of the two different Leica antennas was setto 0.0628(m). This value was obtained using antenna calibration information availablefrom the National Geodetic Survey at http://www.ngs.noaa.gov/ANTCAL/. The phasedifference of the L2 signal was used, in preference to the L1 signal due to the fact thatSeptentrio has indicated that their receivers only initiate an RTK ambiguity resolutionin the case a L2 signal is available. However most likely Septentrio uses both L1 andL2 in an integral way for the final position estimate. This has been neglected in thisevaluation.

Table 5.7: Static performance results.

position, heading and pitch error RTK fixed

Bias Standard Deviation (1σ) 95th percentile

X position error (m) -0.0053 0.0043 0.0108Y position error (m) -0.0015 0.0018 0.0047Z position error (m) -0.0054 0.0035 0.0110Heading error (deg) -0.0447 0.0425 0.1105

Pitch error (deg) -0.0852 0.0660 0.1896

In the table 5.7 basic statistical data of the RTK fixed position error is presented.Bias is obtained by comparing the mean of the data set to the marker point positions,and are supplemented by confidence intervals, marking the region within which the


(a) No-plate dynamic solution

(b) HPPC dynamic solution

Figure 5.12: Comparison of the dynamic solutions.


(c) Carbon fibre dynamic solution

Figure 5.12: Comparison of the dynamic solutions.


(a) High elevation satellite

(b) descending satellite

Figure 5.13: C/N0 (signal strength) ratios of two satellites under three circumstances.


Figure 5.14: Skyplot over Delft, during the static experiment.


true bias falls with 95% certainty. Standard deviations are given as well and are againaccompanied by confidence intervals. Finally the 95th percentiles are also given andrepresent the distance to the median within which 95 percent of position estimationsare represented.

The results show complete compliance with both the requirements set by the Su-perbus, as the information provided by the manufacturer, as provided in section 5.2.The 1σ accuracy values are far smaller the mandatory minimum, showing that the sys-tem is theoretically able to provide the necessary precision to be used for applicationssuch as an active suspension, described in section 5.2. The static nature of this test, aswell as the close proximity to the base station and the omission of the UMTS connec-tion do provide a basis for optimum results. It does however show the potential of thecurrent set-up.

A t-test of the data revealed that the hypothesis of a zero mean could be rejected atthe 5% significance level for both the single antenna position, as for the dual antennaheading and pitch. These biases can be categorised small, as the bias and standarddeviation values combined still comply to the requirements. That said, the biases aremost likely a result of satellite geometry, for they are not constant in time, as can beseen in figure 5.15. Atmospheric effects should not be present in this very short base-line.

Figure 5.15: X position and heading error over time.

Lastly, a short note is given on the 3D speed estimates of the Septentrio receiverin static condition. As became clear from section 3.6 it can be expected that these


estimations are very good under normal circumstances. Indeed this is the case in thestatic test. A standard deviation of 0.0156 m/s is calculated, and the 95th percentile is0.0608 m/s. Figure 5.16 shows the errors in speed estimation during the test.

Figure 5.16: Speed error over time.

From this test it can be concluded that in optimal conditions, the GPS system de-signed for the Superbus will conform to the necessary requirements.

Static test, reacquisition

The goal of this experiment was to gain an understanding of the reacquisition be-haviour of the GPS receiver, after a loss of lock of all, or at least a significant numberof tracked satellites has occurred. This has been described in 5.3. The parametersrelevant for GPS performance are explained in section 5.1, the experimental set-up insection 5.3

The data constitutes 2223 epochs, measured every second, with 56 losses of lockduring the test. Availability is not applicable here, since satellites were intentionallyblocked. However during the test, the receiver locked up twice, refusing to reacquirean RTK solution, both float or fixed. The receiver was reset and the test was resumedin the same fashion.

The results provided, in table 5.8, are subdivided in the three solution possibilities,stand alone, RTK float and RTK fixed. For all types of solution, the bias, standard


Figure 5.17: Skyplot over Delft, during the reacquisition experiment

deviation and 95 percentile are given. Confidence intervals are supplied where appli-cable.

During the processing of the test data, it became apparent that the receiver had es-timated and fixed the ambiguities incorrectly during one reacquisition cycle. This canbe seen clearly in figure 5.18 with a RTK fixed position approximately 2.7 (m) off inx position and about 2.0 (m) in y position. Because of the rather large role this wrongambiguity fix plays in the statistical variables, the relevant reacquisition iteration hasbeen omitted from the data set to acquire the values in table 5.8.

Figure 5.18: Demonstration of incorrect RTK ambiguity fix.

Note that the position solutions are worse than the pure static test, but when an


RTK fixed solution is available, the results are still very good, and compliant to therequirements. The difference is partly misleading and can be partly explained. A de-teriorating RTK fix solution is still given, in the few seconds after the signal is beingblocked. This is clearly visible in figure 5.20 around the 1900 epoch, where a diver-gent RTK fixed trail is present before the RTK fix is ultimately lost. For the stand aloneand RTK float solutions it can be mentioned that the conditions are the worst possibleduring the reacquisition procedure, with the receiver trying to reacquire signals whilstsimultaneous computing solutions, and shifting to a higher graded solution type, thefirst moment this is possible. Therefore care should be taken in judging these results,only limited epochs are present, and those present have a higher variance. Confidenceintervals are therefore typically larger then in the previous static case.

Table 5.8: Static performance results of the reacquisition test

Mode Std dev (1σ) 95th percentile

X position error (m) stand alone 1.8788 3.5828RTK float 1.8714 3.4171RTK fixed 0.0097 0.0178

Y position error (m) stand alone 1.5080 3.1272RTK float 0.8030 1.6260RTK fixed 0.0074 0.0083

Z position error (m) stand alone 3.5447 6.9813RTK float 2.4743 4.4189RTK fixed 0.0139 0.0240

Heading error (deg) stand alone 10.3119 8.0225RTK float 19.0006 40.0303RTK fixed 9.3275 30.2785

Pitch error (deg) stand alone 6.5096 23.7425RTK float 35.3474 66.2149RTK fixed 16.2679 51.1518

Heading and pitch solutions fare worse, this can be explained twofold. Again thesignal reacquisition plays a role, also it is found that the second antenna may not yethave an RTK fixed solution, when the (global) fixed flag is raised within the receiver,see figure 5.19. Note the convergent trails at the 1700 and 1800 epochs (The trail at the1900 epoch is due to signal blockage). This is invisible for the end user, and gravelyinfluences the statistical results.

The fixing success rate can be calculated in two ways, either scoring the two lock-ups as failures or discarding them from the calculation. In the first case the successrate yields 94.8% ± 5.7% in the latter 98.2% ± 3.45%. The 95% confidence intervalsaccompanying these figures are calculated by equation 5.2, where p is the ratio offailures and N the sample size. Both fixing success rate results have intervals upholdingthe Septentrio claim of 99%. Note that a 100% fixing success rate will not be possible,measurements are always imperfect, and therefore a small chance of an incorrect fix


will always remain.

ci = 1.96

√p(1− p)

N(5.2)

Figure 5.19: Heading error due to unfixed secondary antenna.

Figure 5.20: Detail: Divergent heading error before loss of RTK.


The results of the reacquisition times are shown in figure 5.21. Here one can see aclear minimum of 19 seconds. A second peak is visible at 24 seconds and a third at 29,which supports the theory that an interval of 5 seconds is maintained to attempt a RTKinitialisation. Additionally a clear decreasing trend can be seen in reacquisition times.With such a clear distinction of 5 second intervals, the smaller peaks adjoining thelarger ones could be presumed to have the same reacquisition length as the large peak.Figure 5.22 shows the same information, but represented in cumulative percentages.Here one can see that nearly 95% of reacquisitions are within 30 seconds, the meanbeing 22.73 seconds.

Figure 5.21: Histogram of the reacquisition times. A total of 56 trials

Dynamic test, Duifpolder

The goal of the Duifpolder test was to assess receiver performance in preferred con-ditions. A large open area with little multipath and ample satellites in view. This testcould be used to validate the system performance, additionally it functions as a base-line for the subsequent tests in more challenging environments. This goal could becompleted with the obtained data, albeit with some adversities.

Since the Duifpolder is located in the centre of a large rural part between Delft andRotterdam, UMTS service was erratic. The connection was lost several times duringthe experiment. In total 3061 epochs were recorded, of which 1333 were RTK fixed.1701 were stand alone while 27 were RTK float. Float solutions were only used forinitialisation, after radio failures. Under normal operation the receiver could maintainan RTK fix solution. Conditions were good and sunny, while the number of satellitesin view ranged from 5 to 8. Note that only 15.8 percent of the epochs feature less than8 satellites, while only 1 percent features less than 7.


Figure 5.22: Cumulative representation of reacquisition times.

Figure 5.23: Track of the Duifpolder experiment. [source: Google earth]

Although the circumstances were not ideal from the UMTS perspective, the exper-iment was a success. In part, even due to the erratic UMTS reception. In addition to thenormal performance figures, obtained with RTK, some additional information could beextracted from the data. Section 5.3 made it clear that it is nigh impossible to obtain afaultless ground truth for a dynamic test, while still maintaining unrestricted mobility.The ground truth is obtained using post processing, and provides precision figures inthe order of centimetres. The performance of the Septentrio RTK solution howeveris judged by comparing post processed observations to the Septentrio obtained RTKsolution. Therefore, the performance figures provided in these dynamic test are depen-


Figure 5.24: Skyplot over Delft, during the Duifpolder experiment.


dent upon the precision of both receivers.

The issue now for the dynamic test is the following: due to the method of per-formance calculations, it is not possible to quantify the standard deviations of eachpart of the system individually. In other words, only the effective error between thesolutions and the combined standard deviation are available, and we are therefore un-able to comment on performance of the Septentrio receiver individually. To clarify:in order simplify calculations, a fixed point on the vehicle was taken as a referencefor a local North, East, Up, reference frame conversion. This point, the rear Trimblereceiver (Trimble B), was always deemed [0 0 0 ]. This allows for an easier creation ofthe virtual Septentrio position from both Trimble receivers for a moving vehicle, andallows an easier comparison between the two solutions to obtain the effective error.The downside is now that no comments can be given on the individual Trimble andSeptentrio standard deviations, since these two are already linked a priori.

Fortunately static observations were a byproduct of the Duifpolder test. The staticintermissions between up and down legs can provide us with information regardingthe individual receivers, and hence can help us to comment on the quality of the ob-tained results. Below the results of a 100 second static measurement is given for threecases. These three cases will serve as an example to indicate the quality of the resultsobtained during the kinematic tests.

Table 5.9: Static standard deviations (1σ) for the solutions on the dynamic platform

Standard deviation North (m) East (m) Up (m)

Trimble A 0.0039 0.0035 0.0078Trimble B 0.0033 0.0030 0.0055Trimble combined observations 0.0018 0.0013 0.0044

In table 5.9 three sets of standard deviations are given of the 100 second staticmeasurement. For both the first and the second Trimble receivers individually and inaddition also the combined standard deviations of the two Trimbles are given. Theobservations for the latter are obtained by subtracting the difference to the (data set)mean of both Trimble receivers for every epoch, and taking the difference. This isrepresented in equation 5.3.

zi = (xA,i− µA)− (xB,i− µ

B) (5.3)

Now observe that the results of the individual Trimble receivers and the combinedobservation standard deviation following from 5.3 do not coincide. The standard de-viation of the combined observations is actually smaller than that of an individualreceiver. This is an important result. It shows that the precision of combined obser-vations is better than an individual solution. Now remember that the performance ob-servations for the Septentrio receiver are similarly dependent upon multiple receivers,


and the significance becomes apparent. It may well be that the obtained Septentrioperformance results may appear better than they actually are.

This in itself is not a problem as long as the results can be explained, and that isindeed the case. If the results from table 5.9 are correct, this means that observationsfrom both Trimble receivers, although independent, are positively correlated. To verifythis, we will assess the variance and covariance propagation for linear functions [30].If the results are indeed correlated, we can comment on the results of the Septentrioas well. The combined Trimble observations are obtained in a simplified but largelyanalogous way as the Septentrio observations, as explained in 5.3.

Rewriting the right half of equation 5.3 to:

zi = XA−XB (5.4)

for the north, east and up directions respectively , one can set up the linear systemas follows:

z−= Mu− =

1 0 0 −1 0 00 1 0 0 −1 00 0 1 0 0 −1

NAEAUANBEBUB

(5.5)

Using this, the (a posteriori estimated) covariance matrix for the combined Trimbleobservations, Qzz, can be found using the a posteriori estimated covariance matrix ofthe individual Trimble observations, Quu:

Qzz = MQuuMT (5.6)

Quu, is easily estimated by many computational programs from the 100 static ob-servations used for this example. And as expected, it shows a correlation betweenobservations of the two receivers. This is shown below:

Quu =

15.1 −7.5 14.1 11.4 −5.4 11.7−7.5 12.2 −2.3 −4.5 9.6 −15.414.1 −22.8 60.9 8.3 −17.2 35.711.4 −4.2 8.3 10.9 −3.4 8.4−5.4 9.6 −17.2 −3.4 8.7 −12.211.7 −15.4 35.7 8.4 −12.2 30.2

·10−6 (5.7)

Consequently obtaining Qzz and extracting the standard deviations, one acquiresthe following results:

It becomes clear that there is indeed a positive correlation between both Trimblereceivers that causes the joint observations to be more precise than a single solution.This can be explained by the fact that both solutions are relative solutions and use


Table 5.10: propagated standard deviations of both Trimble receivers

standard devation North (m) East (m) Up (m)

0.0018 0.0013 0.0044

a base station on top of the NMI building to obtain the centimetre accurate results.Recall the fact that baseline solutions use the temporal and spatial correlation of theionosphere and tropospheric delays between rover and base station. Now recall thatboth Trimble receivers use the same base station and are placed close to each other inthe field. This means that both Trimble receivers are affected by atmospheric differ-ences between rover and base station in nearly the same way. Both receivers use thesame measurements from the base stations, and receive practically the same signals be-cause of their close proximity to each other. Therefore it becomes easy to understandthat both baseline solutions are correlated, and hence that the combined observationscan be more precise than a single solution: the analogous errors of the both baselinesolutions are cancelled out.

Now let’s look at the behaviour of the individual Septentrio receiver during this100 second static period, and to the combined Septentrio-Trimble observations. Thestandard deviations of both are represented in table 5.11. Once again we see the be-haviour that the combined solution is better than the individual solution. Luckily thiscan be explained now.

Table 5.11: static standard deviations (1σ) for Septentrio and Septentrio-Trimble so-lutions.

Standard deviation North (m) East (m) Up (m)

Septentrio solution 0.0037 0.0050 0.0072Septentrio-Trimble combined solution 0.0032 0.0037 0.0065

Now that this behaviour is known we can cautiously comment on the quality ofthe results of this dynamic test. It is safe to say that the baselines between the vehicleand the NMI building are definitely correlated. The result of this is that a combinationof observations, as used to calculate the Septentrio performance, is affected by this.How much this combination of observations is affected is difficult to say because ofthe static nature of this deduction and the short sample period. It will be dependent,amongst others, on baseline length, atmospheric conditions and receiver type. It seemssafe to say that the performance results provided in this section are equal of slightlybetter in comparison to the individual Septentrio results. Although this might seem aweak deduction, it actually does provide a good qualitative measure considering thestochastic nature of the provided ground truth.

A qualitative base for the dynamic test is now established. But before the actualresults are presented it is important to know how these are represented. Although up


until now, figures were always represented to 10−4 metres, all the following dynamictest results will be presented to 10−3. The main reason for this, is the vehicle an-tenna platform of section 5.3. This was measured to the millimetre level. Furthermore,although this platform is assumed rigid and flat, the dynamic forces and vibrationsduring the test, makes precisions to the tenth of a millimetre useless. Additionally theantenna height of the Septentrio base station was only measured to 10−3. Note thatthis has no influence on the standard deviation figures, yet the 95th percentile resultswill be influenced.

Another comment on the performance figures involves the height. One might no-tice these figures are often worse than the other performance figures. This has in part todo with the GPS system in general, which usually has more difficulty providing preciseheights. But more importantly the platform did not provide any method of calculatingroll angles. Therefore it was assumed flat and local level in the general case: steadilydriving on a flat road. Yet this is not always the case, for example at higher speed turns,potholes and road shoulders. Although up-direction performance figures are presentedfor the dynamic test, it is important to note that these figures are less informative thanthe remainder of the parameters. This is especially notable in the Duifpolder wherethe static RTK results mentioned above were obtained on the road shoulder, resultingin a 2 centimetre bias in up direction.

Continuing with a note on the actual figures that are and are not presented in the ta-ble. First, the standard deviation figures are 1σ values and are presented accompaniedwith an interval. This interval represents the 95% confidence interval of the solution.The intervals are obtained by carrying out a variance test on the data. If this intervalfalls outside the registered precision of the table (below 0.0005), it is represented by0.000. Secondly the 95th percentile figures are presented. This is usually done for allthree solutions types, stand alone, RTK float and RTK fixed. It is decided that biasesare not represented in the table. Under normal RTK fix circumstances all these valueswere very small (10−3), indicating no systemic or methodical errors. This has alreadybeen assessed in section 5.4. Any deviation from this norm indication for example awrong ambiguity fix or high multipath environments will be treated individually, pro-viding more background information.

Lastly note that speed is mentioned nowhere in this table. This is due to the Trimblebaseline solution. This solution provides a position solution with a reasonably restric-tive 1 Hz update rate. More importantly, velocity is not computed in post processing,resulting in the need to numerically differentiate two separate position solutions to ob-tain velocity information. This differentiation over a relatively long interval in turnleads to an inevitable decrease in precision with higher accelerations. The slow updaterate is less of an issue for heading information since instantaneous information of thetwo Trimble receivers can be used. Note that specific information on speed accuracywill be given below, but providing the dry figures in the main performance table wouldpaint a misleading picture of the performance.

In Table 5.12 the results for the Duifpolder experiment are given. Consider thatRTK was never lost unless the UMTS failed, and that the few available float solutions


are only during static initialisation. They are therefore discarded in the table. They donot represent any real aspect of the dynamic precision test.

The table reveals some interesting data. It is easy to see that while in stand alonemode, the position performance goals of section 5.2 are logically not met. A generalrepresentation of the position error in relation to PVT Mode can be found in figure5.25. Yet looking more at the heading yields a more interesting perspective. One cansee that the Septentrio receiver provides a heading precision that is far more precisethan can be expected from two stand alone positions, and one that is in line with anRTK solution. This is confirmed when looking at the RTK heading figures. These nighon coincide. Figure 5.26 shows the heading error of the receiver over the entire Duif-polder experiment. Below it, the PVT mode is presented. One can see that the headingerror of the Septentrio is irrespective of the PVT mode, precision never decreases whenRTK is lost. This is interesting and is caused by the calculation method Septentrio hasimplemented for the secondary antenna. A small baseline is set up, between the mainand the secondary antenna, where the observations of the main antenna are used foran RTK solution of the secondary antenna. This allows for precise heading and pitchdata even while in stand alone mode. Note that heading figures do not suffer the samestandard deviation uncertainties, as discussed above. Both headings can be compareddirectly without the need of additional calculations.

Figure 5.25: Position error in relation to PVT Mode.

The same conservation of precision holds for the speed information. It does notdegrade while in stand alone mode. Although this is to be expected in light of section3.6, confirmation of this is still of importance for Superbus. Looking at figure 5.27more closely also reveals the reason why speeds are not integrated in table 5.12. Thisfigure shows the difference between the numerically differentiated Trimble velocity,


Tabl

e5.

12:D

uifp

olde

rDyn

amic

test

resu

lts.

Solu

tion

Para

met

erN

orth

(m)

Eas

t(m

)U

p(m

)H

eadi

ng(d

eg)

Stan

dA

lone

Stan

dard

devi

atio

n(1

σ)

0.44

9±

0.01

50.

145±

0.00

50.

964±

0.03

20.

097±

0.00

395

thpe

rcen

tile

1.99

50.

519

2.37

80.

217

RT

KFi

xed

Stan

dard

devi

atio

n(1

σ)

0.01

0±

0.00

00.

008±

0.00

00.

034±

0.00

10.

095±

0.00

395

thpe

rcen

tile

0.02

00.

019

0.08

50.

181


Figure 5.26: Heading error in relation to PVTMode and NrSV.

and the Septentrio velocity output directly. In the figure one can see no obvious dis-continuities signalling a precision change when changing PVT Modes, shown in thefigure 5.27. Note also the static periods at the 1500 and 2000 second epochs. Theyshow no obvious difference in noise. This is in line with expectations. More inter-esting to see is the clear sinusoidal nature of the speed error. This is caused by thenumerical differentiation introducing a fault in the comparison. The vehicle’s contin-uous acceleration and deceleration show up as errors in opposite directions.

Another observation is presented in figure 5.28. Here one can see an apparent in-dependence between heading error and speed. This again can be explained through thedifferent ways the receiver obtains these figures. The same observation holds betweenspeed and position error, but this is not shown here.

Now that most observations regarding this experiment are made, it is time to returnto the core table of 5.12 and comment on how these figures fare against the require-ments and manufacturer claims. Logically Stand alone figures do not comply with Su-perbus requirements. Yet heading performance is excellent and complies completelywith the Septentrio claim of 0.10 degrees standard deviation on a baseline of 3 metres.Since the test vehicle baseline was 2.795 metres, performance is in line with manu-facturer claims, even when in stand alone mode. Looking at the RTK fixed figures,shows complete compliance with Superbus requirements, even when reserving someuncertainty due to correlation issues.

Although not all epochs of the Duifpolder test were RTK fixed solutions, this isonly due to technical issues not related to the GPS receiver. The receiver was able tokeep RTK fixed solutions whenever possible and therefore one can conclude that RTK


Figure 5.27: Receiver speed comparison of the Duifpolder experiment.

Figure 5.28: Septentrio speed in relation to heading errors.

availability was 100%, when not hindered by technical difficulties. When in RTKfixed mode it becomes clear that 95th percentile figures are very good, and complywith the requirements. Furthermore the standard deviations are in line with manufac-turer claims, keeping in mind that the baseline was nearly 7 kilometres. Note that thetest conditions, except for UMTS reception, were realistically very good conditionsfor the Superbus: Straight road, good satellite reception and low multipath conditions.Yet it is an important conclusion that the complete GPS system is able to provide the


required performance with ease.

From this section it is easy to conclude that the GPS system as presented in thisthesis, is practically capable to provide performance in line or exceeding Superbusrequirements in all performance fields. Additionally it is shown that performance isin line with manufacturer specifications. However care should be taken to ensure afunctioning data connection. This was visibly the weakest point of the positioningsystem during this rural test. Heading and speed solutions output by the receiver wereobserved to have the same precision in both stand alone and RTK fixed solutions.No correlations were found between vehicle speed and heading or position precision.Lastly a positive correlation between the two independent receiver solutions was ob-served (Trimble and Septentrio). It is shown that it influences the performance figurespresented here. The presented results are however approximately in line with the ob-served performance of the individual Septentrio receiver.

Dynamic test, Emerald

The goal of the Emerald experiment was to evaluate GPS performance in less thenideal circumstances. This urban environment poses worst case conditions during highway travel. This high speed scenario has the most to gain from an high accuracy GPSsystem, since urban uses will mainly entail standard satellite navigation that can stillfunction properly with decreased accuracy. UMTS conditions were good during theexperiment, and the connection remained in operation. The receiver had access to cor-rection data throughout the test. This eliminates an important mitigating factor in theother experiments, where UMTS connections were prone to fail. The receiver couldtherefore work under the intended conditions from the technical side. This allows fora proper evaluation of the intended GPS system in less then ideal real-world conditions.

Figure 5.29: Track of the Emerald experiment. [source:Google Earth]


Figure 5.30: Skyplot over Delft, during the Emerald experiment.

Conditions were good and sunny, while the local environment posed normal tohigh urban multipath conditions. This can be illustrated by the driven track, visible infigure 5.29. The track was chosen to obtain a mix with closed streets, and reasonablyopen main road. The closed streets inhibit a full aerial view, reducing the number ofsatellites in view, and introduce large flat surfaces and objects, which can instigate mul-tipath. The main roads have a larger aerial view, although still less than the selected 10degree cut-off angle, and are less prone to multipath, although some large areas couldstill interfere. A skyplot of the Emerald during the experiment can be found in figure5.30.

The obtained data set contains 1438 epochs, with a maximum of 8 satellites inview. 9 epochs in 3 situations did not result in a PVT solution, 81 resulted in a standalone solution, 99 in a RTK float solution, and finally 1249 resulted in a RTK fixedsolution. Note that epochs for which no solution was found, also register that no satel-lites where in view and occurred in the ’closed street’ sections of the track, in thenorth-west section and the south-east section (while passing a large flat).

The performance figures obtained during the test are interesting, considering thedifficult multipath environment and the fact that one third of the epochs (490) haveonly 6 or less satellites in view, while 174 epochs only have 5 satellites in view. Look-ing at the first plot (figure 5.31), where the 3D position error is given, the error betweenthe ground truth and the Septentrio track, one notices the relatively poor performanceof the solution during the first 1000 seconds of the test even when an RTK fixed solu-tion is available. Initially this is a little over 50 centimetres. It can also be seen thatthere are 2 distinct breaks in trends, one around 470 seconds where the error dropsto about 15 centimetres and one can also be observed around 1130 seconds to nearlyzero. The final epochs of the test show a divergence again. The vehicle and systemsetup remained unchanged during the entire test.

All the observations above give rise to the possibility that the ambiguities were


Figure 5.31: Septentrio error to ground truth in relation to the PVT Mode.

fixed incorrectly during the initialisation of the experiment. The Trimble ground truthdid not show any discontinuities. The fact that the initialisation was performed stat-ically on a parking adjacent to two large buildings supplies some credibility to thishypothesis of wrong ambiguity fixing. In addition in the final epochs of the test, wherethe solution starts to diverge again, the vehicle was parked on the same location. As-suming the ambiguities were incorrectly fixed, some interesting behaviour can be de-duced from the data. Although it cannot be conclusively stated that the ambiguitieswere not correct, looking at the distinct convergent behaviour of the data set towards acorrect solution, does strongly suggest this.

The data shows that the receiver is quite effective in keeping the satellite ambi-guities fixed as long as not all satellites are lost, or just for a very brief time. Thiscan be seen in figure 5.32 where the PVT solution is completely lost, and zero satel-lites were indicated as in view for 2 epochs. The receiver was able to regain the RTKfixed very quickly, within 5 seconds, without resorting to a complete re-initialisation,and losing precision in only two epochs. Even in the few instants the receiver couldnot compute a solution, the receiver likely continued tracking several satellites andretained their ambiguities. This behaviour is beneficial for Superbus use, since it willincrease the robustness of the RTK solution as well as increase the availability of RTK.

The consequence of this behaviour is that the ambiguities will remain incorrect fora longer period of time in difficult environments, just as can be observed in this test.However, the benefits of a robust RTK algorithm far outweighs the re-initialisationscenario, which has become clear in section 5.4. To alleviate this ambiguity fixingbehaviour the RTK algorithm employed by Septentrio clearly does maintain a running


Figure 5.32: Septentrio loss of solution behaviour.

Kalman filter even when ambiguities are fixed, since the solution eventually convergesto the right ambiguities, as observed in figure 5.31. If we assume initially incorrectfixed ambiguities, we must then also observe the behaviour where the solution con-verges, to the correct ambiguities.

The heading error graph of figure 5.33 is also interesting. Even if the ambiguitiesare incorrect, the heading error resulting from this fact, should be constant. This how-ever is not traceable in the figure. The receiver clearly struggles with the combinationof multipath and incorrectly fixed ambiguities, be it from one or both antennas. Theerror is not constant, and shows quite a large amplitude, unfamiliar for an RTK fixedsolution. In some cases when RTK is lost, for example in the 900 to 1000 secondinterval, the heading errors aggravates even more. This leads to the hypothesis that thereceiver has some trouble with the ambiguities for both antennas, due to the multipath.In section 5.4, it has been found that a RTK fixed flag is raised, even if the secondaryantenna is not yet fixed. At the same time section 5.4 showed that under proper con-ditions, the heading error is independent to the PVT Mode. An unfixed secondaryantenna could explain the larger noise than usual for an RTK fixed solution (see figure5.35 when only 5 satellites are visible).

When the receiver has converged to the most accurate solution, from about 1100seconds into the experiment, it is clear that the heading error values are far better thanthe previous sections. Here the adverse conditions of the test track are undeniably ofsignificantly less influence. Heading errors stay within expectations, and are far lessprone to multipath and reduced satellite visibility, see figure 5.36.

In light of the previous paragraphs it is decided that two sets of performance re-sults are formed. One for the complete duration of the test, and one for the last 300seconds of the experiment. Indeed the last section of the test shows significantly betterresults in all aspects, while the test conditions remained the same. The receiver was


Figure 5.33: Heading error in relation to the NrSV and PVT Mode.

Figure 5.34: Heading error in relation to the NrSV and Velocity.

far less prone to lose RTK fixed, position and heading errors were far better, and es-pecially heading errors were far less susceptible to multipath. Significant performancegains are visible, while: the satellite geometry remains largely the same, the test trackremains identical, and the drive speeds are comparable (see figure 5.34). The last 300seconds are therefore deemed an example of a correctly functioning RTK system in anadverse environment and included in the results. Results of the this ’High Accuracy’section will be given twofold. Once for only the RTK fixed solutions of this sec-


Figure 5.35: Heading error detail, showing dependence to the NrSV.

Figure 5.36: Heading error detail, showing dependence to the NrSV.


tion and once for the all solutions combined, Stand Alone, RTK float and RTK fixed.These last results concern the complete technically functioning positioning system, inadverse conditions, without distinction between solutions types. The availability of theRTK fixed solution therefore plays a role in the results. They are therefore included asa benchmark to establish if the system is fit for use in the Superbus.

In tables 5.13 the results of the 2 scenarios are given: the complete test, and thelast 300 seconds. One can clearly see that the results are improved for the High Accu-racy case. Clearly the high accuracy results comply to the 5 centimetre 95th percentilerequirement, with an exception for the height in the total high accuracy scenario. Yeta lower performance can always be expected in the vertical direction. The 95th per-centile values of the complete test are certainly worse than the high accuracy part, butwhen RTK is fixed they are still within decimetre range. This still makes the systemuseful for high precision navigation at low speeds. It is therefore still fit for use for in-ner city navigation. This is exactly the scenario where these high multipath conditionspresent themselves. Yet it may be clear that they are not in compliance to the require-ments. The heading figures do not fit the manufacturer claims. Even when ambiguitiesare fixed, they are approximately a factor 10 higher. This is due to the high multipathsituation, looking at the skyplot of the Delft during the test (figure 5.30) reveals ratherdisastrous conditions, with only 2 satellites above a 45 degree angle. This makes directcontact with most of the satellites difficult in narrow streets, resulting in the potentialuse of the reflected signals.

Concluding this section, one can confirm that the proposed positioning system stillworks in though, high multipath, low sky visibility conditions. It has become clear thatinitialisation of the system should be done with care, in low multipath conditions, toavoid incorrect ambiguity fixing as much as possible. Yet the system showed a rela-tively good robustness in keeping an RTK fixed position in adverse conditions. Evenwhen RTK was lost, it was regained very quickly without complete reinitialisation. Inaddition it has been shown that even if ambiguities were not correct, the receiver showsauto correcting behaviour, jumping to ambiguities that better fit the observations. Eventhough RTK was less precise than the Duifpolder experiment, it still satisfied require-ments. Especially interesting to note is the total high accuracy condition. This denotesthe last 300 seconds of the test, irrespective of PVT mode. This situation somewhatsurprisingly complies to the requirements for north and east directions, and shows thetotal behaviour during adverse conditions. This is deemed very good. It is noted thatsystem performance in all fields improved once the correct ambiguities were fixed, beit position and heading precisions, or robustness of the RTK fixed solution.

Dynamic test, A4

The goal of the A4 test was to evaluate the receiver performance during short black-outs caused by viaducts and the like. While driving under these underpasses, lockswith all available satellites may be lost, causing the need for the reacquisition proce-dure of the RTK solution. Since the Superbus concept leans on flexibility, a wide range


Tabl

e5.

13:E

mer

ald

dyna

mic

test

resu

lts.

Solu

tion

Para

met

erN

orth

(m)

Eas

t(m

)U

p(m

)H

eadi

ng(d

eg)

Stan

dA

lone

Stan

dard

devi

atio

n(1

σ)

2.14

0±

0.33

90.

533±

0.08

41.

946±

0.03

89.

356±

0.00

395

thpe

rcen

tile

4.80

01.

316

4.30

519

.568

RT

KFl

oat

Stan

dard

devi

atio

n(1

σ)

0.92

8±

0.13

20.

344±

0.04

91.

427±

0.20

424

.751±

3.54

895

thpe

rcen

tile

2.30

00.

822

2.05

910

.277

RT

KFi

xed

Stan

dard

devi

atio

n(1

σ)

0,14

2±

0.00

60.

038±

0.00

20.

152±

0.00

62.

174±

0.08

595

thpe

rcen

tile

0.34

80.

074

0.34

73.

795

Hig

hA

ccur

acy,

RT

KFi

xed

Stan

dard

devi

atio

n(1

σ)

0.01

4±

0.00

10.

013±

0.00

10.

031±

0.00

30.

554±

0.04

695

thpe

rcen

tile

0.03

00.

036

0.07

11.

632

Hig

hA

ccur

acy,

Tota

lSt

anda

rdde

viat

ion

(1σ

)0.

071±

0.00

60.

044±

0.00

40.

096±

0.00

80.

573±

0.04

695

thpe

rcen

tile

0.04

60.

047

0.08

11.

635


of operational theatres may be encountered. Underpasses are omnipresent in Dutch in-frastructure and therefore the behaviour in such conditions must be established.

Figure 5.37: Track of the A4 experiment. A Viaduct is present at the bottom of thetrack, an aqueduct crosses the halfway point of the track. [source: Google earth]

Figure 5.38: Skyplot over Delft, during the A4 experiment.

Although the experiment was successfully concluded, conditions were far fromoptimal during the test. The UMTS connection failed numerous times and the con-ditions near the 2 passed viaducts block precious satellites and cause multipath. Thetest yielded 1983 data points, with a maximum of 8 satellites in view. 1166 epochswere RTK fixed, while 77 were float solutions. 698 epochs were stand alone, althoughmost of these are due to a failing UMTS connection. In addition, not all passes yieldeduseful data because the lack of corrections prohibited the receiver to return to the RTKfixed solution. 15 reacquisition procedures could be reconstructed from the data, while


some failed due to UMTS failure.

One curious anomaly presented itself during the test. At around 1200 seconds thereceiver diverted to different ambiguities, while maintaining an RTK fixed solution.This behaviour occurred, not while driving under an underpass, but rather on top ofa viaduct, while changing to the southerly leg of the highway track. This resulted inabout 45 seconds of PVT solutions with incorrect fixed ambiguities, whereafter thereceiver corrected its mistake. The correction itself took 3 seconds, during which thereceiver indicated an RTK float solution (See figure 5.39). This behaviour was earlierencountered in section 5.4, yet here unfortunately ambiguities change from correct toincorrect. Note that no epochs noted a loss of all satellites, and no underpass waspassed. It is possible that the behaviour is caused by the sidewall of the viaduct, inthe 10 to 15 seconds prior to the jump in ambiguities. This wall provides an reason-ably homogenous multipath environment during that time. The behaviour again showsthat the receiver does not stop searching for better ambiguity fits, even when theseambiguities are fixed to integers. In unfortunate instances, in this case probably thecombination of multipath with an emerging satellite, this can lead to a wrong decision.No further unexpected receiver behaviour was found.

Figure 5.39: Total Position error during A4 with respect to PVT Mode.

Returning to the test results, it once again becomes clear in table 5.14 that therequirements are met for north and east directions, when an RTK fixed solution is


present. In up direction is the result approaches requirements very closely. Standalone and RTK float results are expectedly much worse due to the adverse conditions.That RTK float solutions are worse than in the Emerald experiment can be explainedby the fact that needs to recover from a complete satellite black out in a multipath en-vironment, while the receiver could mostly keep track of at least some satellites. Thisallows the receiver to keep the ambiguities fixed for these satellites and regain an RTKfixed solution faster. In the A4 situation, (almost) all satellites are lost and all ambigui-ties must be recalculated. Section 5.4 shows that RTK float solutions need to convergecoming from a pure stand alone solution. This explains the high standard deviations.

The results of passing the viaducts paint a mixed picture. 40 percent of under-passes were passed without a complete loss of PVT solution. This means that thesolution of the next epoch was stand alone, or in one case even RTK float. If the PVTsolution was lost, it was restored, on average, within 3.44 seconds. On average it took11.85 seconds before RTK fixed was restored. This is significantly less than the 22.73seconds of the static test. This is quite impressive, and possibly due to the fact that notall satellite signals are completely lost in this scenario.

Commenting on the reacquisition behaviour of the receiver during the underpassis difficult. The fact that two underpasses were present relatively close to one anotherposes additional difficulties for the receiver. The loss of the data connection whiledriving under these viaducts influences the test even more. This caused some mea-surements to fail. In total 15 successful results were obtained. The cumulative time tofirst fix probability figures of the time it takes the receiver to reacquire a fix are pre-sented in figure 5.40.

Figure 5.40: Cumulative time to first fix probabilities, absolute and RTK fixed.

Concluding this section is a little more difficult than the previous cases. It can bestated that Superbus performance requirements are met for both horizontal directions.The vertical direction error is only slightly greater than allowed. Standard deviationsof the RTK fixed solution are also in compliance with manufacturer specifications.


Tabl

e5.

14:A

4dy

nam

icte

stre

sults

.

Solu

tion

Para

met

erN

orth

(m)

Eas

t(m

)U

p(m

)H

eadi

ng(d

eg)

Stan

dA

lone

Stan

dard

devi

atio

n(1

σ)

0.57

7±

0.03

00.

391±

0.02

11.

254±

0.06

643

.511±

2.37

295

thpe

rcen

tile

1.13

80.

827

2.56

313

.181

RT

KFl

oat

Stan

dard

devi

atio

n(1

σ)

1.59

2±

0.25

90.

448±

0.07

31.

672±

0.27

256

.195±

9.48

095

thpe

rcen

tile

1.13

70.

932

4.18

312

.668

RT

KFi

xed

Stan

dard

devi

atio

n(1

σ)

0.01

3±

0.00

00.

008±

0.00

00.

031±

0.00

12.

674±

0.10

995

thpe

rcen

tile

0.02

00.

019

0.05

20.

238


Figure 5.41: Heading errors versus PVT Mode and Number of satellites in view.

Heading figures are much worse and one may question the usefulness. Several largeoutliers are present in the data, which severely influence the results. Perhaps this isagain due to the behaviour of the secondary antenna. A filter applied to the headingestimates may improve the usefulness the results in these conditions, because gener-ally, as figure 5.41 shows, figures are still quite good. This is confirmed by the 95thpercentile result for RTK fixed. Reacquisition heading performance after a black outhowever is unusable and should probably be discarded for some duration after an un-derpass is encountered. They do not comply with specifications. Again the robustnessof the RTK solution is shown around the 1200 second mark (see figure 5.39), albeitthis time in a negative way.

Chapter 6

Conclusion

In this report a high accuracy positioning system is investigated, designed and evalu-ated for use in the Superbus. Requirements for the system were limited but rigid:

The Superbus Positioning System, must be able to provide real time,precise positioning data. Horizontal position error may not exceed morethan 5 centimetres in 95% of the obtained solutions.

These requirements only leave augmented positioning as a possibility for the Superbus.This is in itself not a problem, but it does put extra strain on the additional requirementof (quasi)national deployment. Real Time Kinematic (RTK) relative positioning wasselected as a valid means to obtain the performance requirements. At the time ofwriting the alternative, Precise Point Positioning was not robust and fast enough forpractical use in the vehicle. The effective area of deployment of RTK can be enlargedsufficiently by using networked base stations and Pseudo Reference Stations PRS. Net-work Transport of RTCM via Internet Protocol (NTRIP) was the selected method ofexchanging the RTCM data, correction data for augmented positioning, over the inter-net, making the system far more flexible than systems using direct data connectionssuch as mobile radio transceivers. UMTS was selected as the communication systemfor the correction data. UMTS will allow for soft hand-overs between the land basedcells as the vehicle moves, allowing for an uninterrupted data stream to the Superbus.In addition, UMTS is able to accommodate the high cruising speed of the Superbusvehicle.

The GPS system in the vehicle consists of two Antennas, spaced 8.19 metres apart,coupled to one Septentrio PolaRx2eh GPS receiver. The receiver is coupled to a com-puter receiving the required correction data from any appropriate base station or basestation network. Using this set up, manufacturer specified performance figures shouldbe within the Superbus requirements.

Several tests were carried out to evaluate real world performance and to verifythe manufacturer specifications, and validate that the system fulfils Superbus require-ments.

135

136 CHAPTER 6. CONCLUSION

A materials test was carried out to investigate the effects of cover plating over theantennas. Results showed that Carbon Fibre plating is highly unfit for this purpose,while the alternative, a thermoplastic composite HPPC has no noticeable effect on po-sition estimation.

A static test evaluates the Superbus RTK system, both for static performance, andposition reacquisition behaviour after a black out situation. The performance resultsshow complete compliance with both the requirements set by the Superbus, as wellas to the information provided by the manufacturer. The reacquisition test was carriedout 56 times. During the test the receiver locked up twice, and once estimated theincorrect ambiguities, although the 99% success rate stated by the manufacturer canbe upheld. From the observations it can be concluded that the secondary antenna usedfor pitch and heading, is not necessarily fixed yet, when the receiver indicates a RTKfix solution. 95% of the reacquisition times were shown to be within 30 seconds, witha 19 second minimum, a 54 second maximum and an average of 22.73 seconds.

Three kinematic tests were carried out, assessing real world behaviour on a movingvehicle. The Duifpolder experiment, provided preferred conditions with ample satel-lites an a low multipath environment. Results showed that Superbus requirements were(logically) not met while in stand alone mode. It must however be noted that headingand speed estimates remained equally precise regardless PVT mode, and compliant tomanufacturer specification. In RTK fixed mode, both Superbus requirements, as man-ufacturer claims were upheld and exceeded by the system.

The goal of the Emerald experiment was to evaluate GPS performance in less thenideal, urban circumstances. The local environment posed normal to high urban mul-tipath conditions. Results showed that the system was able to satisfy requirementsand specifications in these conditions as well, although only in the last 300 secondsof the test. The first part showed a large and fluctuating bias (approx. 20 to 50 cm)in the position estimates as well as poor heading solutions, giving rise to the assump-tion that the system had initialised with incorrect ambiguity estimates. Environmentalconditions at the initialisation area may have allowed for this. The test showed that asa consequence the receiver had more difficulty maintaining an RTK fixed solution inthe first part of the test, while conditions and track remained the same. Yet the systemshowed a relatively good robustness in keeping an RTK fixed position in adverse urbanconditions when ambiguities are estimated correctly. Even when RTK was lost, it wasregained quickly without complete reinitialisation.

Note that conditions were difficult with 46% of the epochs tracking 6 or less satel-lites, and only 2 or 3 satellite above 45 degrees elevation. It was shown that even ifambiguities were not correct, the receiver showed auto-correcting behaviour, jumpingto ambiguities that better fit the observations.

The goal of the last A4 test was to evaluate the receiver performance during shortblack-outs in real world conditions, caused infrastructure such as underpasses. Con-ditions were far from optimal during the test. The UMTS connection failed numeroustimes and the conditions (large noise barrier) near the 2 passed viaducts block precious

137

satellites and cause multipath. 15 scenarios could be reconstructed from the data. In40 percent of these cases, underpasses were passed without a complete loss of PVTsolution. This means that the solution of the next epoch was stand alone, or in one caseeven RTK float. If the PVT solution was lost, it was restored, on average, within 3.44seconds. On average it took 11.85 seconds before RTK fixed was restored

Concluding the A4 test is difficult because of the difficult testing conditions, butit can be stated that Superbus performance requirements are met for both horizontaldirections in RTK fixed mode. The vertical position error is only slightly larger thanallowed. Standard deviations of the RTK fixed solution are also in compliance withmanufacturer specifications. This a surprising, but good result. Heading figures are notin accordance to the specifications. Yet 95th percentile figures are still low at 0.238degrees with RTK fixed, showing that this performance may be improved by applyinga filter to the results. Reacquisition heading results are however best avoided for sometime after a black out.

As a result of the testcampaigns, it can be concluded that the Superbus GPS sys-tem, as it is described it this report, can fulfil all Superbus requirements. Additionally,specifications of the manufacturer are deemed realistic and attainable. It can be con-cluded that the results are good when UMTS is available, even in adverse conditions,making it potentially fit for use in real world conditions.

Chapter 7

Discussion and Recommendations

The conclusion showed that the Superbus GPS system, designed in this report, is ableadhere to all Superbus requirements. This does however not mean that this is uncon-ditionally so. This, together with other observations and recommendations, will bediscussed here.

First the boundary conditions of the GPS system will be clarified:

• The Superbus GPS System only fulfils all requirements with an RTK fixed solu-tion

• A loss of RTK fixed solution can occur in adverse conditions

• Reinitialisation will take time, and incorrect ambiguity fixing can occur (in ad-verse conditions)

• UMTS reception is crucial for the Superbus GPS System

• Antenna placement can have a great impact on performance and system robust-ness

Since the availability of the necessary high accuracy RTK estimates is completely de-pendent on the UMTS connection, it is crucial that this connection is operational. Inany other case, the estimates from the receiver will not be meaningfully better than anylow-cost receiver, and hence not adhere to Superbus requirements, except for speed andheading estimates.

UMTS reception however has been an issue during testing. Connection issueswere observed and should therefore be investigated further. Although low cost UMTSequipment was used, and antenna placement was not optimal, it remains an importantissue. Additional testing should verify connection availability and functionality in highspeed conditions and poor reception areas. Additionally, some form or auto-reconnectscheme should be employed in the case of a lost connection.

During (high accuracy) operation, care should be taken in using the computed esti-mates in critical situations as multipath (both urban and large homogeneous multipath

139

140 CHAPTER 7. DISCUSSION AND RECOMMENDATIONS

conditions such as sounds walls) has been shown to impact estimates. It can be recom-mended that alarm limit (HAL and VAL) values are defined depending on the actualpositioning requirements in various situations. The receiver supports RAIM and theuse of these alarm limits, and can give timely warning when the accuracy of the com-puted solutions cannot be guaranteed to be within these limits.

RTK robustness, and position estimate integrity are very important when using theGPS system for critical systems. Therefore, as a precaution, it can be recommendedthat RTK initialisation is performed in low multipath conditions as much as possible.It is shown in testing that in the event of incorrect ambiguity fixes, the position esti-mates will slowly converge to the ground truth, indicating a movement to the correctinteger ambiguities. It is also made plausible that the robustness of the system andsystem performance is worse in these conditions, hence the recommendation for a lowmultipath RTK initialisation. Although difficult to test unambiguously, it may be rec-ommended to further test the receiver during high multipath conditions and establishif the receiver indeed performs worse if ambiguities are incorrectly fixed, and to eval-uate the convergent behaviour that was observed during the Emerald test. As an addedmeasure to minimise wrong ambiguity resolution and multipath, one might in the fu-ture investigate the use of a database of visibility levels or useful elevation angles onmuch travelled routes and urban areas. Such a database is relatively easy to obtain[45]. The database can warn drivers to adverse conditions depending on location, orperhaps eventually, the use of the information can be integrated in the GPS receiveritself, filtering out or weighting satellites depending on elevation angle, azimuth andvehicle location.

Further recommendations regarding the receiver, aimed more directly to the man-ufacturer, include a warning flag, indicating an unfixed secondary or tertiary antenna.The behaviour which is found now, where it is unknown to the user whether or notthe heading and pitch estimates are RTK fixed, float or stand alone, may be dangerousand undesirable. Furthermore, a locked up receiver as witnessed during the reacquisi-tion test, is not acceptable for Superbus use, as it takes several minutes to reboot thereceiver and reestablish a RTK fixed estimate. Although the lock ups occurred duringtorture testing, and were not observed during the kinematic tests, it may be advisableto test the underpass interruption scenario more thoroughly to evaluate and understandthe behaviour of the receiver better under these conditions. Perhaps a clear strategycan be devised on what receiver information may and may not be used in the momentsafter an interruption.

Lastly some general recommendations. It may be important to stress the potentialbenefits of the GPS system during initial testing. The high performance nature of theGPS system and the controlled conditions available at a test track give the possibility ofdetailed and precise vehicle path reconstruction and can help to observe and establishthe behaviour of the vehicle in both normal and abnormal conditions. The availabilityof attitude information at 10 Hz (course over ground, heading and pitch) can also bevaluable information. Braking and suspension behaviour may be assessed using pitchdata and it may be advantageous to explore the limits of the vehicle grip and assessbehaviour during emergency situations.

141

It is also good to realise that possibilities of the UMTS system in the Superbusexceeds that of the GPS system alone. It enables the development of other systemssuch as fleet management systems or other online location based services.

Finally some practical notes regarding the use of the GPS system as it is installedin the Superbest now. First it is important to feed the antenna position of the secondaryantenna (w.r.t. the main antenna) in the receiver. This can aid time to first fix values,and decrease the probability of a wrong ambiguity fix for this antenna. Secondly it isimportant to evaluate the elevation cut-off angle. It can be important to change thisvalue, depending on the conditions that are present in the roof sections where the an-tennas are installed. Lastly it is important to note that the use of the standard NMEAmessages may prohibit the use of centimetre accurate output. It can be beneficial touse the SBF format instead, which can provide additional information as well.

Bibliography

[1] International Standards, Recommended Practices and Procedures for Air Navi-gation Services - Annex 10, 4 edition, 1985. International Civil Aviation Organi-zation (ICAO).

[2] Aeronautics and Space Engineering Board and Commission on Engineering andTechnical Systems, editors. Summary and Proceedings of a Workshop on Im-proving the GPS Reference Station Infrastructure for Earth, Oceanic, and Atmo-spheric Science Applications. National Academies Press, 1997.

[3] S. Bisnath and Y. Gao. Current state of precise point positioning and futureprospects and limitations. In F. Sans and M. G. Sideris, editors, Observingour Changing Earth, International Association of Geodesy Symposia. SpringerBerlin Heidelberg, 2008.

[4] T. Brown. Handling at the limits. GPS World, 8 2010.

[5] S. N. P. V. Cauwenberge. Study of soft handover in UMTS. Master’s thesis,Technical University of Denmark and University of Gent, Belgium, 7 2003.

[6] O. Colombo and A. Sutter. Evaluation of precise, kinematic GPS point position-ing. Institute Of Navigation, 9 2004. Proceedings of the Institute Of Navigation.

[7] R. Dawkins. The Selfish Gene. pages 238 – 253. Oxford University Press, 1976.

[8] R. Descartes. Discours de la methode pour bien conduire sa raison et chercher laverite dans les sciences. http://www.gutenberg.org/ebooks/13846 (French)or http://www.gutenberg.org/ebooks/59 (English).

[9] R. Descartes. Principia Philosophiae. http://digbijzcoll.library.uu.nl/lees_gfx.php?lang=nl&BoekID=1184 or http://www.gutenberg.org/ebooks/4391 (English).

[10] D. Dettmering, C. Waese, and G. Weber. Networked Transport of RTCM viaInternet Protocol, NTRIP, Version 1.0, Example Implementation. Bundesamt frKartographie und Geodsie (BKG), Frankfurt, Germany, 2 2006.

143

144 BIBLIOGRAPHY

[11] Federal Aviation Administration. TSO-C129a, airborne supplement navigationequipment using the global positioning system (GPS), 2 1996.

[12] J.-P. Henry. State Space modellering binnen een RTK netwerk, 2 2005. Workshopdifferentiele GPS netwerken, NIN/GIN/HSB.

[13] M. Hernandez-Pajares et al. Wide-area RTK, high precision positioning on acontinental scale. pages 35 – 46. Inside GNSS, 3 2010.

[14] S. Hewitson and J. Wang. Gnss receiver autonomous integrity monitoring (raim)performance analysis. GPS Solutions, 10:155–170, 2006.

[15] P. Hwang and R. Brown. From RAIM to NIORAIM. pages 24 – 33. InsideGNSS, 5 2010.

[16] E. Kaplan and C. Hegarty. Understanding GPS Principles and Applications.Artech House Inc., 2 edition, 2006.

[17] H. Landau, U. Vollath, and X. Chen. Virtual Reference Stations versus broadcastsolutions in network RTK - advantages and limitations. GPS World, 5 2003.

[18] R. B. Langley. Dilution of precision. GPS World, 5 1999.

[19] I. Leveson. L2C Study: benefits of the new GPS civil signal. Inside GNSS, 72006.

[20] R. Litjens. The impact of mobility on UMTS network planning. Computer Net-works, 2002.

[21] R. Litjens. HSDPA flow level performance and the impact of terminal mobility.IEEE Wireless Communications and Networking Conference (WCNC), 2005.

[22] A. Lo. Ae4-e07: Location-based services - mobile and wireless networks, 42007. Part of Course, Location-based Services, Delft University of Technology.

[23] R. Loh and J. Fernow. Integrity Monitoring Requirements for FAA’s GPS Wide-area Augmentation System (WAAS). Center for Advanced Aviation System De-velopment, 1994.

[24] G. Mader. GPS antenna calibration at the National Geodetic Survey. NationalGeodetic Survey, NOS/NOAA.

[25] G. McKessock. A comparison of local and wide area GNSS diffential correc-tions disseminated using the network transport of RTCM via internet protocol(NTRIP). 4 2007. Department of Geodesy and Geomatics Engineering, Univer-sity of New Brunswick.

[26] J. Melkert. Veiligheid Superbus. Faculty of Aerospace Engineering, Delft Uni-versity of Technology, The Netherlands, 11 2005.

[27] J. Melkert. Bouwdoos infrastructuur Superbus. Faculty of Aerospace Engineer-ing, Delft University of Technology, The Netherlands, 3 2006.

BIBLIOGRAPHY 145

[28] J. Melkert. Ontwerpverantwoording en systeembeschrijving Superbus. Facultyof Aerospace Engineering, Delft University of Technology, The Netherlands, 32006.

[29] J. Melkert. Superbus: Using aerospace technology to make high speed trans-port more sustainable. Faculty of Aerospace Engineering, Delft University ofTechnology, The Netherlands, 11 2007.

[30] E. Mikhail. Observations and least squares. University Press of America, 1982.

[31] P. Misra and P. Enge. Global Positioning System: Signals, Measurements, andPerformance. Ganga-Jamuna Press, 2 edition, 2006.

[32] W. Y. Ochieng, K. F. Sheridan, K. Sauer, X. Han, P. A. Cross, S. Lannelongue,N. Ammour, and K. Petit. An assessment of the raim performance of a combinedgalileo/gps navigation system using the marginally detectable errors (MDE) al-gorithm. GPS Solutions, 5:42–51, 2002.

[33] D. Odijk, P. de Bakker, and S. Verhagen. Fast LAMBDA-based ambiguity reso-lution combined with WARTK technique for Galileo data simulated using ESA’sGalileo signal validation facility. pages 40 – 52. European Journal of Navigation,11 2004. Presented at the IGNSS Symposium, December 2009.

[34] J. Oliveira and C. Tiberius. Landing: Added assistance to pilots on small aircraftprovided by EGNOS. pages 321–333. IEEE/ION, 2008. Position, Location andNavigation Symposium, 2008 IEEE/ION.

[35] T. Oudemans. Echte filosofie. Uitgeverij Bert Bakker, 2007.

[36] S. Pullen, T. Walter, and P. Enge. System overview, recent developments, and fu-ture outlook for WAAS and LAAS. Department of Aeronautics and Astronautics,Stanford University, 11 2002.

[37] Radio Technical Commission for Maritime Services. RTCM Recommended Stan-dards for Differential GNSS (Global Navigation Sattelite Systems) Service, Ver-sion 2.3 RTCM Standard with Amendment 1, 5 2010. www.rtcm.org.

[38] R. Sabatini and G. B. Palmerini. Differential global positioning system (DGPS)for flight testing. Technical report, Research and Technology Organisation, 2008.RTO AG-160 Vol. 21 / SCI-135.

[39] J. Schiller. Mobile Communications, second edition. Pearson Education Limited,2007.

[40] Septentrio Satellite Navigation NV. How to retrieve NTRIP data using GNSSInternet Radio.

[41] Septentrio Satellite Navigation NV. PolaRx2/2e User Manual, version 3.2.0, 12007.

[42] Septentrio Satellite Navigation NV. Create An RTK Setup Using Two PolaRx2eReceivers, 2008.

146 BIBLIOGRAPHY

[43] C. Shankwitz. Elbow room on the shoulder. GPS World, 7 2010.

[44] A. Simsky and F. Boon. Positioning algorithms for the new polarx2 receiver.2003. 11th IAIN World Congress.

[45] M. Takahashi. A novel prediction method for vehicle and satellite interactionin the simulation with GPS signals collected. In K. Koyamada, S. Tamura, andO. Ono, editors, Systems Modeling and Simulation, pages 262–266. SpringerJapan, 2007.

[46] P. Teunissen. An optimality property of the integer least-squares estimator. pages587 – 593. European Journal of Navigation, 1999.

[47] P. J. G. Teunissen. A new method for fast carrier phase ambiguity estimation.pages 562 – 573. IEEE/ION, 1994. Position Location and Navigation Sympo-sium, 1994.

[48] P. J. G. Teunissen. The least-squares ambiguity decorrelation adjustment: amethod for fast GPS integer ambiguity estimation. Journal of Geodesy, 70, 1995.

[49] P. J. G. Teunissen and S. Verhagen. The GNSS ambiguity ratio-test revisited: Abetter way of using it. pages 138 – 151. Survey Review, 4 2009.

[50] Trimble Terrasat GMBH. Support of Network Formats by Trimble GPSNET Net-work RTK Solution, 6 2005.

[51] R. van Bree, C. Tiberius, and A. Hauschild. Real time satellite clocks in singlefrequency Precise Point Positioning. pages 2400 – 2414. IEEE/ION, 9 2009.Institute of Navigation Symposium, ION GNSS 2009.

[52] T. Verduyckt and J. Melkert. Bijdragen voor de strategische milieubeoordelingvan de Superbus. Faculty of Aerospace Engineering, Delft University of Tech-nology, The Netherlands, 1 2006.

[53] S. Verhagen and P. Joosten. Analysis of integer ambiguity resolution algorithms.pages 40 – 52. European Journal of Navigation, 11 2004.

[54] U. Vollath and H. Landau. Network RTK versus single base RTK - understandingthe error characteristics. 8 2002. presented at ION GPS 2002.

[55] G. Weber, D. Dettmering, and H. Gebhard. Networked transport of RTCM viainternet protocol (NTRIP). volume 128. 2005. Proceedings of the 18th Interna-tional Technical Meeting of the Satellite Division of The Institute of Navigation(ION GNSS 2005).

[56] G. Wubbena, A. Bagge, and M. Schmitz. Network-based techniques for RTKapplications. 11 2001. Presented at the GPS Symposium, GPS JIN 2001.

[57] G. Wubbena, A. Bagge, and M. Schmitz. PPP-RTK: Precise point positioningusing state-space representation in RTK networks. 9 2005. Presented at the 18thInternational Technical Meeting, ION GNSS-05.

BIBLIOGRAPHY 147

[58] G. Wubbena and S. Willgalis. State space approach for precise real time posi-tioning in GPS reference networks. 6 2001. Proceedings of the InternationalSymposium on Kinematic Systems in Geodesy, Geomatics and Navigation.

Acronyms

C/A Coarse Acquisition codeCDMA Code division multiple accessCOTS Commercial-Of-The-ShelfCRC Cyclic Redundancy Check

DoD Department of Defense

EDGE Enhanced Data Rates for GSM Evolution

FDMA Frequency Division Multiple AccessFMS Fleet Management System

GMSK Gaussian minimum-shift keyingGNSS Global Navigation Satellite SystemGPRS General Packet Radio ServiceGPS Global Positionint SystemGPST GPS TimeGSM Global System for Mobile Communication

HAL Horizontal Alarm LimitHDOP Horizontal Dilution of PrecisionHERL Horizontal External Reliability LevelHPPC High Performance thermoPlastic Composite

ICAO International Civil Aviation OrganizationIGS International GNSS ServiceINS Inertial Navigation System

L1 1575.42 MHz GPS satellite signalL2 1227.60 MHz GPS satellite signalLAMBDA Least-squares AMBiguity Decorrelation Adjust-

ment method

149

150 ACRONYMS

LBS Location Based Services

MDB Minimal Detectable BiasMGP Mathematical Geodesy and Positioning group

NGS National Geodetic SurveyNMEA The National Marine Electronics AssociationNMI Nederlands MeetInstituutNrSV Number of Satellites in ViewNTRIP Networked Transport of RTCM via Internet Proto-

col

P(Y) Encrypted precision codePDOP Position Dilution of PrecisionPfa Probability of false alarmPmd Probability of missed detectionppm Parts per millionPPP Precise Point PositioningPRN Pseudo Random NoisePRS Pseudo Reference StationPVT Position, Velocity, Time

RAIM Receiver Autonomous Integrity MonitoringRDW RijksDienst WegverkeerRF Radio FrequencyRINEX Receiver Independent Exchange FormatRL Reliability LevelRTCM Radio Technical Commission for Maritime ser-

vicesRTK Real Time Kinematic

SBAS Satellite Based Augmentation SystemSBF Septentrio Binary FormatSPS Standard Positioning Service

TDMA Time division multiple accessTDOP Time Dilution of PrecisionTNO Nederlandse organisatie voor Toegepast-

Natuurwetenschappelijk OnderzoekTOW Time Of Week

UHF Ultra High FrequencyUMTS Universal Mobile Telecommunications SystemUTC Universal Time, Coordinated

VAL Vertical Alarm LimitVERL Vertical External Reliability Level

ACRONYMS 151

VHF Very High FrequencyVRS Virtual Reference Station

Appendices

153

Appendix A

Septentrio PolaRx2e Specifications

155

PolaRx2e@/PolaRx2eH The PolaRx2eH and PolaRx2e@ are the new generation Heading and Multi-antenna receivers in the PolaRx2platform. Implemented on a single Euro-card size board, the receivers address a wide range of precise headingor attitude positioning and navigation applications in fields like machine guidance, marine surveying andphotogrammetry. Designed for tough field conditions, the PolaRx2eH and PolaRx2e@ are built aroundSeptentrio’s advanced GNSS chipset and offer high quality, high update rate positioning, heading, pitch and roleinformation.

Precise Heading and Attitude at 10 Hz The PolaRx2eH and PolaRx2e@ receivers bring the quality and flexibility of the PolaRx2 platform, and output with high accuracy, the position and velocity at update rates of up to 10 Hz to the field of heading and attitude based applications. The PolaRx2eH heading receiver is the latest variant in the PolaRx2e family and can be connected to 2 dual-frequency antennas to output accurate heading & pitch or heading & roll information. Its high precision and compact form design make the PolaRx2eH perfectly suitable for machine guidance solutions in agriculture and construction, as well as in marine surveying. The PolaRx2e@ on the other hand, collects and outputs GPS data from up to 3 antennas simultaneously (heading, pitch and roll). It can also output relative positioning of 2 or 3 antennas, which can be used for steering an independently moving part, such as for agricultural and towed equipment. As such, the PolaRx2e@ forms a perfect solution for attitude determination and other multi-antenna applications

Unique Single-Board platform

PolaRx2eH and PolaRx2e@ are implemented on a single Euro-card size board. This lightweight and compact form design, together with flexibility and affordability bring important improvements to traditional GNSS-based heading and attitude applications, whilst conjunctively opening the door for new types of applications. They can be combined with RTK positioning on the same board, offering a unique combination of high precision positioning and attitude solutions.

PolaRx2eH can be connected to 2 dual-frequency antennas whereas PolaRx2e@ can be connected to up to 3 antennas, of which the main antenna can be dual-frequency while the auxiliary antennas are single-frequency. Both receiver types have 48 hardware channels, which can be flexibly assigned to track satellites in single or dual-frequency on 1, 2 or 3 antennas in parallel (i.e. without antenna multiplexing). One or more channels can also track the L1 signal of up to 6 SBAS satellites. Next to rigid antenna set-ups, the receivers can also be used in situations where the relative positions of the antennas are not fixed. The receiver will then calculate and output relative positions precisely.

Superior GNSS technology platform

The precise accuracies and update rates available in PolaRx2eH and PolaRx2e@ receiver are made possible through the same high quality architecture used in all products of the PolaRx2 platform: it is built around Septentrio’s GreFE front-end and GreCo GPS/SBAS baseband processor chips. Very low-noise Doppler measurements are the key to exceptionally precise velocities and position accuracies.

Both variants have the high tracking sensitivity and stability of phase tracking of all Septentrio receivers, allowing users to track more satellites for a longer period of time, even under adverse conditions. The receivers also incorporate Septentrio’s mitigation technique APME, unique in its ability to tackle short-delay multipath.

Flexible Integration Options

The PolaRx2e Heading and Attitude variants are available as a standard Euro -card size board, ensuring easy integration. For ready-to-use solutions, they come in a waterproof IP65 rugged enclosure with sturdy connectors, allowing usage in tough and remote environments. The enclosed receiver offers 4 serial ports, a possibility of 256MB non-removable Compact Flash memory card and Ethernet access. New features include logging control via push button or external signals and programmable LEDs. The intuitive Graphical User Interface program, RxControl, accompanies the variants. RxControl can be used with the receivers for configuration, for logging and remote control and includes advanced visualization possibilities. Possible data output formats are the industry standard NMEA format as well as a compact Septentrio owned binary format.

FEATURES • 48 hardware channels for “all in view”

GPS+SBAS parallel tracking

• All channels configurable to track satellites in single or dual-frequency on 1, 2 or 3 antennas in parallel (i.e. without antenna multiplexing)

• Dual frequency L1/L2 code/carrier tracking

• Includes SBAS channels (EGNOS, WAAS, other)

• Raw data output (code, carrier, SBAS navigation data)

• Up to 10 Hz raw measurement, position and attitude output rate (user selectable)

• Automatic or manual antenna calibration

• A Posteriori Multipath Estimator technique (APME)

• Differential GPS (rover)

•

• output

x PPS output (x = 1, 2, 5, 10)

10 MHz reference input /

• RAIM module included

• Four bi-directional serial ports (RS232), ps baudrate up to 115 kb

• NMEA v2.30 output

Highly compact and detailed• Septentrio

• -ng status and position fix

• nd stop Data output/Logging on

• or mounted in IP65 waterproof

• detailed operating and installation manual

TI

station

• RT

Network compatible (FKP)

• removable Compact

TCP/IP over Ethernet

Binary Format (SBF) output

6 LEDs for power, logging, LAN link, Multipurpose, trackiidentification

Start aEvent

• Compact single-board Euro card solution

OEM boardenclosure

• Sturdy connectors

Includes intuitive GUI (RxControl) and

OP ONS

• Differential GPS base

K (main antenna)

RTCM v2.2, 2.3 or 3.0 input/output

Reference

CMR 2.0

• 2 Event markers

On Board Logging (nonFlash Memory Card)

• Programmable LEDs

•

E

PERFORMANCE Position accuracy1,2

Horizontal3 Vertical3

Standalone 1.1 m 1.9 m SBAS 0.7 m 1.2 m DGPS RTK4,5

0.6 m 1 cm + 1 ppm

1.1 m 2 cm + 2 ppm

Velocity Accuracy1,2

Horizontal3 Vertical3

Standalone 1.5 mm/sec 1,9 mm/sec Attitude Accuracy1,2,14,16

1 m antenna separation Heading 0.3° Pitch/Roll 0.6° 3 m antenna separation Heading 0.1° Pitch/Roll 0.2° 10 m antenna separation Heading 0.03° Pitch/Roll 0.06° Auxiliary Antenna positions15

0.6 mm

Maximum Update rate 10 Hz Latency < 50 msec 1 PPS accuracy1,2 10 nsec Measurement precision1,3,6

C/A pseudoranges7 0.15 m (GPS)8

0.30 m (GPS)9

0.35 m (SBAS) P1/P2 pseudoranges7 0.1 m L1 carrier phase 0.2 mm L2 carrier phase 1 mm L1/L2 doppler 2.5 mHz (0.5 mm/sec) Time to first fix Cold start10 < 90 sec Warm start11 After power-on < 55 sec After reset < 20 sec Re-acquisition Time to first heading/ attitude output

< 2 sec 45 sec

Tracking performance (C/N0 threshold)12,13

Code phase tracking 19 dB-Hz Carrier phase tracking 26 dB-Hz Acquisition 33 dB-Hz Acceleration 4 g Jerk 3 g/sec

1 1 Hz measurement rate

2 Performance depends on environmental conditions

3 1σ level

4 Fixed ambiguities

5 Baseline < 20 km

6 C/N0 = 45 dB-Hz

7 non-smoothed

8 Multipath mitigation disabled

9 Multipath mitigation enabled

10 No information available (no almanacs, no approximate position)

11 Almanacs and approximate position known, no ephemeris known

12 95%

13 Max speed 515 m/sec, max altitude 18 000 m

14 Attitude accuracy increases linearly with antenna separation

15 No multipath

16 PolaRx2eH only Heading and Pitch

POLARX2 H/@ TECHNICAL SPECIFICATIONS PHYSICAL AND ENVIRONMENTAL

Size 160 x 100 x 13 mm (OEM board)

285 x 140 x 37 mm (In housing)

Weight 120 g (OEM board) 930 g (In housing)

Input voltage 5 VDC ± 5% (OEM board)

9-30 VDC (In housing)

Antenna LNA Power Output Output voltage Maximum current

+ 5VDC 200 mA

Power consumption 5 W typical, 7W max Operating temperature -30 to +70 °C Storage temperature -40 to +85 °C Humidity 5% to 95% (non condensing) Connectors

Antenna 10 MHz in PPS out

OEM board Backplane Extension

Housing Power COM1 COM2 OUT/COM3&4 IN Ethernet

TNC female BNC female BNC female

DIN 41612 type B,

64 pins male (consult Septentrio)

ODU 3 pins female ODU 7 pins female ODU 7 pins female ODU 5 pins female ODU 7 pins female ODU 4 pins female

POLARX2E FAMILY : OTHER PRODUCTS

PolaRx2e and PolaRx2e_OEM – PolaRx2e is a versatile dual-frequency GNSS receiver platform for high-end applications. Based on code and carrier tracking of the L1 and L2 signals, it provides the user with satellite range measurements and position, velocity and time. PolaRx2e_SBAS – The single-frequency variant tracks up to 6 SBAS augmentation satellites (such as EGNOS and WAAS) in addition to GPS satellites, offering vital integrity information for application in safety-critical environments. PolaRx2C – The PolaRx2C can track up to 4 satellites in L2C mode. For these satellites, the CA, P1, P2 and L2C measurements are available simultaneously. PolaNt – A lightweight precise positioning and survey single or dual-frequency antenna for use with PolaRx family. RxControl - RxControl is an intuitive user interface to configure and control all types of PolaRx receivers and monitor, log and post data remotely. RxMobile - A unique intuitive, portable GUI field controller for the PolaRx receivers. RxMobile allows controlling the receiver, monitoring the navigation solution and accessing its functions in the field in the same intuitive way as with RxControl. SSNDS 03/2006/2

Appendix B

Septentrio PolaNt Specifications

159

PERFORMANCE Frequency 1227 ± 10 MHz 1575 ± 10 MHz Polarization RHCP Axial Ratio 3dB max

Radiation Coverage

4.0 dBic θ = 0° -1.0 dBic 0° < θ < 75° -2.5 dBic 75° ≤ θ < 80° -4.5 dBic 80° ≤ θ < 85° -7.5 dBic θ = 90°

Amplifier Gain 38 ± 2 dB Noise Figure 2.5 dB max Input Voltage + 5 to +18 VDC Current 50 mA (typ) Impedance 50 Ω VSWR ≤ 2.0:1 Band Rejection 35 dB @ 1675 MHz

PHYSICAL AND ENVIRONMENTAL

Finish Weatherable polymer Weight ≈ 425 gr Diameter 178 mm Connector TNCF Altitude ≤ 6 000 m (20 000 ft) Temperature -40°C to +70°C

POLARX2 FAMILY : OTHER PRODUCTS PolaRx2 and PolaRx2OEM - PolaRx2 is a versatile dual frequency GNSS receiver platform for high-end applications. Based on code and carrier tracking of the L1 and L2 signals, it provides the user with satellite range measurements and position, velocity and time. PolaRx2 SBAS - PolaRx2 SBAS can track up to 6 SBAS augmentation satellites (such as EGNOS and WAAS) in addition to GPS satellites, increasing the accuracy of the position and offering your application vital integrity information increasing the confidence in the position solution for appli-cation in safety-critical environments. PolaRx2@ - PolaRx2@ is a unique single-board dual frequency receiver that can be connected to up to 3 antennas, bringing heading/attitude and other multi-antenna applications within economic and practical reach. PolaRx2TR - PolaRx2TR (Timing/Reference) combines world-class performance in terms of measurement noise, sensitivity and tracking stabi-lity with user-oriented features such as Ethernet communication. PolaRx2TR also provides specific GPS timing functions (1PPS in and out). RxControl - RxControl is an intuitive user interface to configure and control all types of PolaRx2 receivers and monitor, log and post data remotely.

PolaNt is a lightweight high precision geodetic dual-frequency antenna for use with the PolaRx2 family of high performance dual-frequency GNSS receivers. This high-gain antenna incorporates low-noise amplifiers, and is built into a rugged and environmentally sealed housing.

PolaNt

Appendix C

Antenna radiation pattern

161

Gain @ Boresight = + 5.0dBic

At2775-42

Gain @ Boresight = + 6.0dBic

At2775-42

Appendix D

Example output Position Module

165

lat lon alt vn ve vu gndspd cog

52.05846169 4.36053914 -7.056 0.001 0.043 -0.045 0.222 -20000000000

52.05846173 4.36053918 -7.067 0.061 0.028 -0.041 0.283 -20000000000

52.05846177 4.36053919 -7.07 0.038 -0.026 -0.002 0.164 -20000000000

52.05846179 4.36053921 -7.064 0.017 0.021 0.054 0.215 -20000000000

52.05846179 4.36053921 -7.064 0.017 0.021 0.054 0.215 -20000000000

52.05846207 4.36053922 -7.16 0.028 0.044 0.104 0.419 -20000000000

52.05846207 4.36053934 -7.143 -0.015 0.09 0.149 0.63 12.05

52.05846206 4.36053942 -7.131 -0.006 0.018 0.16 0.581 -20000000000

52.05846205 4.36053938 -7.143 -0.045 -0.01 -0.031 0.2 -20000000000

52.05846203 4.36053935 -7.142 0.001 -0.046 0.025 0.189 -20000000000

52.05846203 4.36053939 -7.12 0.039 0.033 0.169 0.635 -20000000000

52.05846206 4.36053931 -7.135 -0.007 -0.07 -0.031 0.276 -20000000000

52.05846205 4.36053927 -7.137 -0.02 -0.004 -0.016 0.093 -20000000000

52.05846201 4.36053924 -7.15 -0.083 -0.036 -0.091 0.463 350.086

52.05846195 4.36053922 -7.157 -0.033 -0.01 -0.053 0.228 -20000000000

52.05846194 4.3605393 -7.14 0.042 0.046 0.128 0.511 -20000000000

52.05846196 4.36053933 -7.134 -0.002 -0.004 0.118 0.424 -20000000000

52.05846196 4.36053926 -7.15 -0.049 -0.026 -0.068 0.316 -20000000000

52.05846194 4.36053931 -7.135 0.049 0.017 0.09 0.373 -20000000000

52.05846196 4.36053929 -7.143 -0.026 -0.016 0.022 0.137 -20000000000

52.05846193 4.36053932 -7.14 -0.045 0.049 0.088 0.398 -20000000000

52.05846188 4.36053935 -7.125 -0.013 -0.047 0.105 0.417 -20000000000

52.05846188 4.3605393 -7.119 0.02 -0.027 0.098 0.371 -20000000000

52.05846191 4.36053928 -7.111 0.034 -0.01 0.138 0.514 -20000000000

52.05846191 4.36053923 -7.127 -0.079 -0.007 -0.066 0.373 200.489

NrSv PVTer PVTMo av corr-age pdop tdop hdop vdop

6 0 1 65535 534 399 184 501

6 0 1 65535 534 399 184 501

6 0 1 65535 534 399 184 501

6 0 1 65535 534 399 184 501

6 0 1 65535 534 399 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 501

6 0 1 65535 534 400 184 502

6 0 1 65535 534 400 184 502

6 0 1 65535 534 400 184 502

6 0 1 65535 534 400 184 502

6 0 1 65535 534 400 184 502

6 0 1 65535 534 400 184 502

6 0 1 65535 535 400 184 502

6 0 1 65535 535 400 184 502

6 0 1 65535 535 400 184 502

6 0 1 65535 535 400 184 502

TOW year month day hour min sec Int

243752100 9 8 11 19 42 17 0

243752200 9 8 11 19 42 17 0

243752300 9 8 11 19 42 17 0

243752400 9 8 11 19 42 17 0

243752400 9 8 11 19 42 17 0

243753400 9 8 11 19 42 17 0

243753500 9 8 11 19 42 17 0

243753600 9 8 11 19 42 17 0

243753700 9 8 11 19 42 17 0

243753800 9 8 11 19 42 17 0

243753900 9 8 11 19 42 17 0

243754000 9 8 11 19 42 17 0

243754100 9 8 11 19 42 19 0

243754200 9 8 11 19 42 19 0

243754300 9 8 11 19 42 19 0

243754400 9 8 11 19 42 19 0

243754500 9 8 11 19 42 19 0

243754600 9 8 11 19 42 19 0

243754700 9 8 11 19 42 19 0

243754800 9 8 11 19 42 19 0

243754900 9 8 11 19 42 19 0

243755100 9 8 11 19 42 20 0

243755200 9 8 11 19 42 20 0

243755300 9 8 11 19 42 20 0

243755400 9 8 11 19 42 20 0

herl-p verl-p herl-v verl-p disbase heading pitch ageoflastcorr

0.278393 0.236439 0.528551 0.799737 -20000000000 -20000000000 -20000000000 0

0.278428 0.236447 0.52855 0.799731 -20000000000 -20000000000 -20000000000 -20000000000

0.278443 0.236459 0.528548 0.799725 -20000000000 -20000000000 -20000000000 -20000000000

0.278469 0.236469 0.528547 0.799718 -20000000000 -20000000000 -20000000000 -20000000000

0.278469 0.236469 0.528547 0.799718 -20000000000 -20000000000 -20000000000 -20000000000

0.277942 0.236397 0.528534 0.799655 -20000000000 -20000000000 -20000000000 -20000000000

0.277901 0.236376 0.528532 0.799648 -20000000000 -20000000000 -20000000000 -20000000000

0.277858 0.236355 0.528531 0.799642 -20000000000 -20000000000 -20000000000 -20000000000

0.277876 0.236339 0.52853 0.799635 -20000000000 -20000000000 -20000000000 -20000000000

0.277858 0.23632 0.528528 0.799629 -20000000000 -20000000000 -20000000000 -20000000000

0.277772 0.236299 0.528527 0.799622 -20000000000 -20000000000 -20000000000 -20000000000

0.277696 0.236276 0.528526 0.799616 -20000000000 -20000000000 -20000000000 -20000000000

0.277699 0.236252 0.528524 0.799609 -20000000000 -20000000000 -20000000000 0

0.277715 0.236234 0.528523 0.799603 -20000000000 -20000000000 -20000000000 -20000000000

0.277834 0.236221 0.528522 0.799596 -20000000000 -20000000000 -20000000000 -20000000000

0.277921 0.236206 0.52852 0.79959 -20000000000 -20000000000 -20000000000 -20000000000

0.277856 0.236187 0.528519 0.799583 -20000000000 -20000000000 -20000000000 -20000000000

0.27789 0.236164 0.528518 0.799577 -20000000000 -20000000000 -20000000000 -20000000000

0.277906 0.23615 0.528516 0.79957 -20000000000 -20000000000 -20000000000 -20000000000

0.277931 0.236145 0.528515 0.799564 -20000000000 -20000000000 -20000000000 -20000000000

0.277884 0.23613 0.528514 0.799557 -20000000000 -20000000000 -20000000000 -20000000000

0.277788 0.236105 0.528511 0.799544 -20000000000 -20000000000 -20000000000 0

0.277653 0.236078 0.52851 0.799538 -20000000000 -20000000000 -20000000000 -20000000000

0.277684 0.236052 0.528508 0.799531 -20000000000 -20000000000 -20000000000 -20000000000

0.277656 0.236027 0.528507 0.799524 -20000000000 -20000000000 -20000000000 -20000000000

Appendix E

Angular velocity precisionestimation

To obtain angular velocities from the Doppler speed calculations of a two antenna re-ceiver, one would get a situation as in figure E.1. The following assumptions are madein the scenario. First the receiver is able to output Doppler speeds for both antennasand secondly, the centre of rotation is assumed on an arbitrary point between the 2antennas. In this way, since the baseline of the two antennas is fixed, the angular ve-locities can be calculated by establishing the relative speed between the two.

Figure E.1: Angular velocities as experienced by the Superbus.

Septentrio supplies horizontal velocity accuracy values of 1.5 (mm/s) in horizon-tal direction (1σ at 1Hz output rate). To get the rotational speed around the centre ofrotation, when the vehicle endures both translational as rotational speeds, one must

169

170 APPENDIX E. ANGULAR VELOCITY PRECISION ESTIMATION

subtract the two velocity vectors from each other. Since the antenna baseline is fixed,this will only leave the rotational velocity. However, both velocity vectors have a stan-dard deviation. If we assume no correlation between the two standard deviations, theresulting combined standard deviation (1σ) of the angular velocities will be approxi-mately 2.12(mm/s). This is shown in equation (E.1).

σc =√(σ1)2 +(σ2)2 = 2.12(mm/s) (E.1)

With the distance from antenna to centre of rotation estimated at 4.1 metres, thiswill give a standard deviation for rotational velocity of 0.0304 (deg/s). One can see inequation (E.2) that the distance to the antenna (d), is a variable, and can be adapted toprevailing conditions.

σΨ=

σc

d=

0.002124.1

= 5.2 ·10−4(rad/s) (E.2)

Documents

Superbus Positioning System · Hiermee wordt duidelijk dat een replicator (genoom) een ratio heeft boven die van het individu (replicant). Je kunt niet anders dan zeggen dat het genoom